# You Might Not Need Frontier

> A source-backed field guide that maps 98 curated prompt examples to four escalating routes, with current model options, explicit exceptions, and verification advice.

- **Reviewed:** 2026-07-25
- **Author:** Sam Carlton — https://samcarlton.com/
- **Canonical guide:** https://youmightnotneedfrontier.com/

## Short answer

For bounded, repeatable work, use deterministic software when the answer is exact. Otherwise start with a local/open or efficient hosted model, verify the result, and move to a balanced or frontier model only when ambiguity, stakes, weak rollback, or weak verification justify it.

The three model names on each example follow a conservative route policy: efficient hosted cards add a Standard-class cushion above a Direct baseline, balanced hosted cards use Deep-class headroom, local cards stay self-hosted, and tool-first cards keep models optional. This is a product heuristic reviewed against SamRouter—not a benchmark, measured sufficiency score, or claim that model tiers are equivalent.

## Guide facts

- 98 curated prompt examples
- 4 route types
- 17 current model records
- 8 reviewed sources
- 6 signals for keeping frontier capability
- Reviewed 2026-07-25

## How to use this guide

1. Match your prompt to the closest example below.
2. Try the first model, or pin a provider/model you already use.
3. Run the listed verifier.
4. Keep or escalate to frontier capability when the exception applies.
5. For a prompt-specific recommendation, use SamRouter: https://router.samcarlton.com/

## When frontier earns its keep

- Ambiguous root-cause debugging across systems
- Open-ended research that must find and reconcile current sources
- Architecture, strategy, or difficult tradeoffs
- High-stakes legal, medical, security, or financial judgment
- Destructive or production-changing work with weak rollback
- Long-horizon agentic work with changing plans and many tools

## Token efficiency and model routing

Model routing and token efficiency are different. This guide minimizes verified task cost through routing; it does not claim smaller models always use fewer tokens. Measure prompt, retrieval, tool-call, retry, verifier, and human-review cost together.

OpenAI reports directional internal coding-agent results where leaner prompt configurations improved scores while reducing total tokens, but the size of the gain varied by workload. Treat that as a reason to run your own evals, not a universal savings promise: https://developers.openai.com/api/docs/guides/prompt-guidance-gpt-5p6

## Model policy

- **Policy:** one class of headroom above the curated route baseline, capped at Deep.
- **Tool-first exception:** models are optional helpers; deterministic software remains the recommendation.
- **Local exception:** the shortlist stays self-hosted and starts with Gemma 4.
- **Input compatibility:** image-consuming cards use image-capable models; local document extraction has an explicit Gemma/Llama/Qwen override.
- **Default order:** OpenAI, Anthropic, then Google when their assigned class and hosting route fit.
- **Pinning:** a provider preference chooses the closest compatible model. An exact-model pin can exceed the suggested class, but it moves first only on cards with a compatible hosted/local mode.
- **Evidence boundary:** SamRouter’s current evals test routing policy on a small synthetic fixture. They do not execute these candidate models or establish minimum sufficient models.
- **Reviewed:** 2026-07-25, against SamRouter policy commit 58e1340054a3982c1a82e19e7a05bc577e758025.

## Full model registry

- [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) — OpenAI; Direct. Efficient, high-volume, well-scoped work.
- [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) — OpenAI; Standard. Balanced intelligence, latency, and cost.
- [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) — OpenAI; Deep. Frontier capability for difficult professional work.
- [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) — Anthropic; Direct. Fast, high-volume, straightforward work.
- [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) — Anthropic; Standard. A speed and intelligence balance for coding and agents.
- [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) — Anthropic; Deep. Accuracy-first complex reasoning and advanced coding.
- [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) — Google; Direct. Fast, low-cost extraction, parsing, and execution.
- [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) — Google; Standard. Complex agentic and multimodal work.
- [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) — Google; Deep · Preview. Preview model for advanced problem solving and multi-step tool use.
- [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) — Google; Direct · Self-hosted. Small multimodal instruction model for on-device use.
- [gpt-oss-20b](https://openai.com/open-models/) — OpenAI; Standard · Self-hosted. Local text reasoning when the runtime and task evals fit.
- [gpt-oss-120b](https://openai.com/open-models/) — OpenAI; Deep · Self-hosted. Larger local reasoning candidate; hardware fit is required.
- [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) — Meta; Standard · Self-hosted. Long-context multimodal work; runtime capabilities vary.
- [Mistral Small 4](https://huggingface.co/mistralai/Mistral-Small-4-119B-2603) — Mistral; Standard · Self-hosted. Sparse multimodal reasoning and function-calling candidate.
- [Qwen3.6 35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) — Qwen; Standard · Self-hosted. Efficient multimodal, thinking, and tool-use candidate.
- [DeepSeek V4 Flash](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash) — DeepSeek; Standard · Self-hosted. Economical long-context reasoning and tools candidate.
- [DeepSeek V4 Pro](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro) — DeepSeek; Deep · Self-hosted. Complex coding and agent workloads on substantial hardware.

## Prompt examples

### 01. Count, sort, deduplicate, or calculate

> Count these rows, deduplicate by email, and calculate the totals.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use SQL, a spreadsheet formula, or a short script. Ask a model to write the operation only if you need help.
- **Why:** The answer is deterministic. A language model adds variability where ordinary code can be exact and repeatable.
- **Keep frontier when:** The hard part is deciding what the numbers mean, reconciling ambiguous definitions, or designing the analysis—not executing the arithmetic.
- **Verify it:** Re-run the operation from the source data; include row counts and invariants.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#count-sort-dedupe

### 02. Extract fields into a schema

> Pull the names, dates, totals, and invoice number from this document as JSON.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta) · [Qwen3.6 35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) (Qwen)
- **Recommendation:** Start local or efficient hosted with a strict output schema. Escalate only for difficult layouts or domain judgment.
- **Why:** The input and required fields bound the job, and schema validation gives you a cheap, objective first check.
- **Keep frontier when:** Scans are poor, tables span pages, fields require legal/accounting interpretation, or missing values have costly consequences.
- **Verify it:** Validate the schema, require source spans or page references, and sample against the original.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-to-schema

### 03. Classify, tag, or route incoming text

> Label each ticket by topic, urgency, and destination team.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a small local or efficient model against a fixed label set. Let low confidence abstain.
- **Why:** Classification is bounded and high-volume—the shape where efficient models and task-specific tuning are strongest.
- **Keep frontier when:** Labels are subjective, policy-sensitive, adversarial, or trigger irreversible action without human review.
- **Verify it:** Measure precision and recall on held-out, human-labeled examples; monitor abstentions and drift.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#classify-tag-route

### 04. Translate routine text

> Translate these support replies into Spanish and return only the translated text.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient hosted model—or a tested local model for privacy and offline work.
- **Why:** Official efficient-model guidance explicitly includes high-volume translation with constrained output.
- **Keep frontier when:** The text is legal, medical, literary, dialect-heavy, safety-critical, or must preserve subtle brand voice.
- **Verify it:** Back-translate a sample, preserve named terms with a glossary, and use native-speaker review for consequential copy.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Google](https://huggingface.co/google/gemma-4-E4B-it)
- **Canonical entry:** https://youmightnotneedfrontier.com/#routine-translation

### 05. Transcribe clear audio

> Transcribe this meeting recording with speaker labels and timestamps.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Start with a local speech model or an efficient multimodal model. Keep the audio private when that matters.
- **Why:** Transcription is a specialized, bounded transform; current efficient and on-device model guides name it directly.
- **Keep frontier when:** Audio is noisy, speakers overlap, vocabulary is specialized, or a verbatim record has legal or safety significance.
- **Verify it:** Spot-check names, numbers, speaker changes, and low-confidence segments against the audio.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Google](https://huggingface.co/google/gemma-4-E4B-it)
- **Canonical entry:** https://youmightnotneedfrontier.com/#clean-audio-transcription

### 06. Rewrite, proofread, or change tone

> Make this email clearer and friendlier without changing the facts.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local or efficient model with explicit facts to preserve and a short style constraint.
- **Why:** On-device guidance demonstrates rewriting and voice-driven editing; the task does not require frontier discovery.
- **Keep frontier when:** The copy carries legal commitments, crisis communications, sensitive negotiation, or a distinctive creative voice that needs expert judgment.
- **Verify it:** Diff facts, names, numbers, dates, promises, and links—not just grammar.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#rewrite-proofread-tone

### 07. Summarize material you provide

> Summarize this report in five bullets for an executive reader.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use local for privacy or efficient hosted for speed. Specify the audience, length, and facts that must survive.
- **Why:** Document processing and summarization are named efficient-model tasks; the supplied source bounds the answer.
- **Keep frontier when:** Synthesis spans conflicting sources, requires current external research, or omission could materially change a decision.
- **Verify it:** Require page or section references and compare every decision-relevant claim with the source.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Google](https://huggingface.co/google/gemma-4-E4B-it) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#summarize-provided-material

### 08. Turn notes into action items

> Convert these meeting notes into decisions, owners, due dates, and open questions.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use local or efficient hosted. Tell it never to invent an owner or date; mark missing fields as unknown.
- **Why:** This is structured extraction plus light summarization, with a clear abstention rule for information the notes do not contain.
- **Keep frontier when:** The meeting is politically sensitive, commitments are disputed, or deciding the real priority requires context outside the notes.
- **Verify it:** Link each item to a quote or timestamp and leave absent owners/dates explicitly unknown.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#notes-to-actions

### 09. Answer from supplied context

> Using only this handbook, answer the question and cite the relevant section.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with retrieval, citations, and an explicit “not in the source” response.
- **Why:** Good retrieval narrows the reasoning surface. The model should explain the supplied material, not invent missing policy.
- **Keep frontier when:** The corpus is huge or contradictory, the question needs cross-document inference, or the answer drives high-stakes action.
- **Verify it:** Open every cited passage; test unanswerable questions and conflicting-source cases.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#answer-from-context

### 10. Draft alt text or image metadata

> Write concise alt text for these product images and tag the visible attributes.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient multimodal model for the first pass, then human-review context and purpose.
- **Why:** The visual input is bounded and the output short. The expensive part is often review, not frontier reasoning.
- **Keep frontier when:** Images are medically, legally, or technically specialized, or subtle relationships are essential to accessibility.
- **Verify it:** Review with the page context; confirm the text conveys purpose rather than listing every visible object.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#alt-text-image-metadata

### 11. Generate known-pattern boilerplate

> Create three components that match this existing component and its tests.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model when the pattern, files, constraints, and check command are explicit.
- **Why:** The repository supplies the pattern and deterministic tools can catch formatting, type, and test failures.
- **Keep frontier when:** The framework is unfamiliar, APIs are uncertain, architecture is changing, or generated code crosses security boundaries.
- **Verify it:** Run formatter, typecheck, tests, and a focused review for copied assumptions.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#known-pattern-boilerplate

### 12. Make a small, scoped code edit

> Rename this API field in the named files, update tests, and change nothing else.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Start balanced; step down to efficient when the edit is mechanical and the verifier is strong.
- **Why:** A clear boundary plus executable checks does more for safe downgrading than prompt length or a model leaderboard score.
- **Keep frontier when:** The task is ambiguous root-cause debugging, a cross-system migration, security-sensitive, destructive, or long-horizon autonomous work.
- **Verify it:** Inspect the diff, run the narrow test suite, and escalate on failure rather than asking the cheap model to improvise indefinitely.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Stanford University](https://arxiv.org/abs/2305.05176) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#small-scoped-code-edit

### 13. Normalize date formats

> Convert every date in this CSV to ISO 8601 and flag invalid values.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a date parser in a spreadsheet, SQL query, or script with an explicit timezone and invalid-value rule.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#normalize-dates

### 14. Rename files from a pattern

> Rename these files to YYYY-MM-DD-title without changing their extensions.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a shell, file-renaming utility, or short script. Preview the rename map before applying it.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#rename-files

### 15. Validate emails, URLs, or identifiers

> Check which rows contain malformed email addresses or URLs.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a standards-aware validator and return the failing row plus reason; use a model only to explain the rule.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#validate-identifiers

### 16. Compare two lists

> Tell me which customer IDs are only in list A, only in list B, or in both.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a set operation, spreadsheet lookup, or SQL join and preserve counts for all three groups.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#compare-lists

### 17. Replace a known text pattern

> Remove the tracking parameters from every URL in this file.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a URL parser or tested regular expression, then diff the result. Ask a model for the pattern only if needed.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#find-replace-pattern

### 18. Join tabular files

> Merge these CSV files by order ID and keep unmatched rows.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use SQL, a dataframe, or a spreadsheet join with an explicit cardinality and unmatched-row report.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#join-tabular-files

### 19. Pivot or aggregate records

> Group sales by month and region, then calculate totals and averages.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a pivot table, SQL GROUP BY, or dataframe aggregation so the calculation stays reproducible.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#pivot-aggregate

### 20. Chart known data

> Make a line chart of weekly signups from this table.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use the charting feature in your spreadsheet or analytics tool; let a model suggest chart types, not invent points.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#chart-known-data

### 21. Find missing required fields

> Find records missing a name, consent date, or source URL.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use schema validation or explicit null checks and report missing fields by row.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#required-field-check

### 22. Compare two text versions

> Show exactly what changed between these two policy drafts.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a word-aware diff tool for the exact changes; use a model separately if you need a plain-language summary.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#diff-text-versions

### 23. Resize or compress images

> Resize these images to 1600 pixels wide and keep each under 300 KB.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use an image processor with fixed dimensions, quality limits, and metadata rules.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#resize-compress-images

### 24. Convert a known file format

> Convert these Markdown files to HTML using the same template.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a deterministic converter such as Pandoc or the project’s existing build pipeline.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#convert-file-format

### 25. Extract contact details

> Pull the names, companies, emails, and phone numbers from these notes.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model with a strict schema when the notes are private; let missing fields remain null.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-contact-details

### 26. Parse receipts

> Extract the merchant, date, tax, tip, and total from each receipt.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient vision model with a receipt schema and confidence or source-box output.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#parse-receipts

### 27. Locate contract clauses

> Find the termination, renewal, and data-retention clauses and cite their pages.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model to locate and quote clauses, then send interpretation to qualified counsel.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#locate-contract-clauses

### 28. Extract product attributes

> Extract size, color, material, compatibility, and warranty from these listings.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with an allowed-value schema and an explicit unknown state.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-product-attributes

### 29. Turn a page into a record

> Turn each event page into JSON with title, date, venue, price, and source URL.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use structured page data first, then an efficient model for fields the markup does not expose.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#webpage-to-record

### 30. Convert a table image to CSV

> Convert this screenshot of a table into CSV without guessing unreadable cells.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use OCR or an efficient vision model with a tabular schema and an explicit unreadable marker.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#table-image-to-csv

### 31. Transcribe a scanned page

> Transcribe this scanned form and mark text you cannot read.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use specialized OCR first, with an efficient vision model for layout recovery and uncertainty labels.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#ocr-degraded-scan

### 32. Detect language

> Label the language of each message and return an ISO language code.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a compact language detector or local model, with unknown for short or mixed-language text.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#detect-language

### 33. Tag sentiment

> Label each review positive, neutral, mixed, or negative.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a small model against the fixed labels and calibrate it on examples from your own domain.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#tag-sentiment

### 34. Flag likely spam or toxicity

> Flag comments that may be spam, harassment, or threats for moderator review.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a tested local classifier as a review queue, not an automatic punishment system; let uncertain cases pass to humans.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#flag-spam-toxicity

### 35. Group text by topic

> Group these feedback comments into recurring themes and name each cluster.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use embeddings or a local model for candidate clusters, then have a human merge, split, and name them.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#cluster-topics

### 36. Score items against a rubric

> Score each request against this published urgency rubric and explain the evidence.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model only for a reversible first-pass queue, with evidence and abstention on unclear cases.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#prioritize-by-rubric

### 37. Detect possible sensitive data

> Flag possible personal, financial, or health information in these documents.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Combine deterministic pattern detectors with a local model for contextual candidates, then require security review.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#detect-sensitive-data

### 38. Review routine community comments

> Apply these community guidelines and send uncertain comments to a moderator.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Start balanced with quoted policy evidence, conservative abstention, and human review for removals or sanctions.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#moderate-routine-comments

### 39. Find duplicate requests

> Find support tickets that describe the same underlying problem.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use local embeddings to propose duplicate groups, while retaining each original ticket for review.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#find-duplicate-tickets

### 40. Route incoming email

> Route each email to billing, technical support, sales, abuse, or other.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local or efficient classifier with a fixed label set, confidence threshold, and other queue.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#route-incoming-email

### 41. Code survey responses

> Apply this codebook to each open-ended survey response.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model against the frozen codebook, allowing multiple labels and an unclear state.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#code-survey-responses

### 42. Classify inbound leads

> Tag each inbound lead by product interest and stated timeline.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model to extract stated intent, not to infer protected traits or automatically deny service.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#classify-sales-leads

### 43. Label feedback themes

> Tag each comment with one or more themes from this taxonomy.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model with the supplied taxonomy and let it suggest new themes in a separate review field.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#label-feedback-themes

### 44. Extract stated resume skills

> Extract only the skills and certifications explicitly stated in these resumes.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model for evidence-linked extraction only; keep hiring judgment and protected-trait controls with people.
- **Why:** A constrained first pass can reduce review labor, but the decision remains sensitive enough to require abstention and human ownership.
- **Keep frontier when:** The output directly determines access, employment, safety, compliance, money, or another consequential outcome.
- **Verify it:** Use representative labeled cases, audit disparate errors, retain source evidence, and keep a qualified human in the decision.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-resume-skills

### 45. Tag a content library

> Tag each article by format, audience, topic, and funnel stage.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model with controlled vocabularies and review low-confidence or multi-topic items.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#tag-content-library

### 46. Label product-review aspects

> Tag which reviews discuss fit, durability, shipping, price, or support.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient multi-label classifier and evaluate each aspect separately on real reviews.
- **Why:** A fixed label set turns an open-ended language task into a bounded decision that smaller models can be evaluated on.
- **Keep frontier when:** Labels are subjective, adversarial, policy-sensitive, or trigger a costly action without human review.
- **Verify it:** Test precision, recall, and abstentions on held-out human-labeled examples, including rare and ambiguous cases.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#label-review-aspects

### 47. Shorten supplied copy

> Cut this announcement to 100 words without removing dates or links.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model with a hard length limit and an explicit list of facts that must survive.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#shorten-copy

### 48. Expand shorthand notes

> Turn these shorthand notes into complete sentences without adding facts.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model and instruct it to mark ambiguous fragments instead of completing them imaginatively.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#expand-shorthand-notes

### 49. Fix grammar and punctuation

> Correct grammar and punctuation but keep my wording and paragraph breaks.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local grammar tool or model with a minimal-edit instruction and review the diff.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#fix-grammar-punctuation

### 50. Apply a style guide

> Edit this copy to follow our capitalization, number, and terminology rules.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use deterministic lint rules where possible, with a local model for the contextual remainder.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#apply-style-guide

### 51. Draft headline variants

> Give me ten accurate headline options for this article.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model to generate candidates from the article, then have an editor select and revise.
- **Why:** The model is producing candidates from supplied facts and constraints; selection and factual review remain human work.
- **Keep frontier when:** The campaign needs original strategy, sensitive positioning, regulated claims, or a distinctive creative concept rather than variants.
- **Verify it:** Reject invented claims, compare every fact with the brief, and have the responsible editor select and revise the final.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-headline-variants

### 52. Draft email subject lines

> Write eight subject lines for this email without adding claims or urgency.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model for variants, preserving the brief’s facts and banning unsupported scarcity.
- **Why:** The model is producing candidates from supplied facts and constraints; selection and factual review remain human work.
- **Keep frontier when:** The campaign needs original strategy, sensitive positioning, regulated claims, or a distinctive creative concept rather than variants.
- **Verify it:** Reject invented claims, compare every fact with the brief, and have the responsible editor select and revise the final.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-subject-lines

### 53. Rewrite material as an FAQ

> Turn this product guide into a concise FAQ using only the supplied facts.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model, require a source section for each answer, and omit unsupported questions.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#rewrite-as-faq

### 54. Change reading level

> Rewrite this for an eighth-grade reader without changing the meaning.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model and a readability checker, then review domain terms that cannot safely be simplified.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#change-reading-level

### 55. Create a plain-language version

> Explain this policy in plain language and keep every requirement.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model for the draft, but retain the original policy as authoritative and cross-check every requirement.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#plain-language-policy

### 56. Draft social captions

> Write platform-specific captions for this event using only these details.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model for variants, then check dates, venue, links, accessibility text, and platform constraints.
- **Why:** The model is producing candidates from supplied facts and constraints; selection and factual review remain human work.
- **Keep frontier when:** The campaign needs original strategy, sensitive positioning, regulated claims, or a distinctive creative concept rather than variants.
- **Verify it:** Reject invented claims, compare every fact with the brief, and have the responsible editor select and revise the final.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-social-captions

### 57. Draft a product description

> Write a product description from these verified specifications and nothing else.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with the verified fact sheet and a hard ban on inferred benefits or certifications.
- **Why:** The model is producing candidates from supplied facts and constraints; selection and factual review remain human work.
- **Keep frontier when:** The campaign needs original strategy, sensitive positioning, regulated claims, or a distinctive creative concept rather than variants.
- **Verify it:** Reject invented claims, compare every fact with the brief, and have the responsible editor select and revise the final.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-product-description

### 58. Draft a routine support reply

> Draft a friendly reply using this approved refund policy and the customer’s order facts.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model grounded in the policy and order record; require approval before sending.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-support-reply

### 59. Format text as Markdown

> Format this draft as Markdown with headings, lists, and working links.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model or deterministic formatter, then render the result and run a link check.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#format-as-markdown

### 60. Convert prose to bullets

> Turn this paragraph into scannable bullets without changing any claims.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model for the structural rewrite and compare claims sentence by sentence.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#prose-to-bullets

### 61. Personalize outreach from known facts

> Personalize this outreach email using only the facts in this contact record.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model with strict source fields, privacy limits, and human approval before sending.
- **Why:** The model is producing candidates from supplied facts and constraints; selection and factual review remain human work.
- **Keep frontier when:** The campaign needs original strategy, sensitive positioning, regulated claims, or a distinctive creative concept rather than variants.
- **Verify it:** Reject invented claims, compare every fact with the brief, and have the responsible editor select and revise the final.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#personalize-outreach

### 62. Draft a meeting agenda

> Turn these goals and open questions into a 30-minute meeting agenda.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model to sequence the supplied topics, with named outcomes and realistic time boxes.
- **Why:** The task is a bounded conversion from supplied operational material into a known structure or checklist.
- **Keep frontier when:** Priorities are disputed, missing context changes commitments, or the output authorizes an irreversible external action.
- **Verify it:** Trace each output item to the source, leave missing information unknown, and require a human approval before acting.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#draft-meeting-agenda

### 63. Summarize an email thread

> Summarize this email thread into decisions, disagreements, and unanswered questions.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with sender and message references, leaving disputed points explicitly disputed.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#summarize-email-thread

### 64. Summarize support history

> Summarize this customer’s support history, attempted fixes, and current blocker.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model grounded only in the ticket history, with ticket IDs or dates for each key point.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#summarize-support-history

### 65. Draft a changelog from commits

> Turn these merged commits into a user-facing changelog.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model after filtering the real commit set, then verify every claimed behavior in the diff or release.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#changelog-from-commits

### 66. Explain document changes

> Summarize the material changes between these two contract drafts.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a deterministic redline plus a balanced model for a cited summary; keep legal interpretation with counsel.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#summarize-document-diff

### 67. Digest supplied research abstracts

> Compare these paper abstracts by question, method, sample, and reported result.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model with a comparison schema and paper-level citations; do not infer beyond the supplied abstracts.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#digest-research-abstracts

### 68. Answer from product documentation

> Answer this setup question from our documentation and cite the exact page.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with retrieval and a not-found response; keep the linked documentation visible.
- **Why:** Retrieval narrows the answer to supplied evidence, so the model can explain a corpus instead of inventing missing facts.
- **Keep frontier when:** The corpus is contradictory or very large, the question needs difficult cross-document inference, or the answer drives high-stakes action.
- **Verify it:** Open each citation, test unanswerable questions, and require an explicit not-found response when the corpus is silent.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#answer-product-docs

### 69. Build a glossary from supplied text

> Create a glossary of the terms defined in this handbook.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model to extract term-definition pairs with section references, omitting terms the handbook does not define.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#build-glossary

### 70. Create study flashcards

> Turn these notes into question-and-answer flashcards.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model grounded in the notes, and include a source heading or page for each answer.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#create-flashcards

### 71. Create a quiz from notes

> Create a ten-question quiz using only this study guide.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model and require answer keys with source references; discard ambiguous questions.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#create-study-quiz

### 72. Collect decisions across documents

> Find every decision in these meeting notes and link it to its source.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model with document and line references, keeping conflicting decisions separate.
- **Why:** The task is a bounded conversion from supplied operational material into a known structure or checklist.
- **Keep frontier when:** Priorities are disputed, missing context changes commitments, or the output authorizes an irreversible external action.
- **Verify it:** Trace each output item to the source, leave missing information unknown, and require a human approval before acting.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-cross-doc-decisions

### 73. Find contradictions in supplied documents

> Find places where these two policy documents disagree and quote both passages.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model with retrieval, paired citations, and an uncertain category for merely different wording.
- **Why:** Retrieval narrows the answer to supplied evidence, so the model can explain a corpus instead of inventing missing facts.
- **Keep frontier when:** The corpus is contradictory or very large, the question needs difficult cross-document inference, or the answer drives high-stakes action.
- **Verify it:** Open each citation, test unanswerable questions, and require an explicit not-found response when the corpus is silent.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#find-document-contradictions

### 74. Summarize survey comments

> Summarize the recurring themes in these survey comments and include representative row IDs.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model after a reproducible sampling or clustering step, preserving minority and dissenting themes.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#synthesize-survey-comments

### 75. Explain a bounded error log

> Explain this error log and list the three most likely checks to run next.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Start balanced when the log and system context are supplied; demand diagnostic commands before any fix.
- **Why:** The task still benefits from strong reasoning, but a tight scope and executable verifier can make a balanced model the economical first pass.
- **Keep frontier when:** Root cause is ambiguous across systems, rollback is weak, production data is at risk, or repeated checks keep failing.
- **Verify it:** Review the plan and diff, run the narrowest useful checks, and escalate on failed evidence instead of allowing improvisation.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Stanford University](https://arxiv.org/abs/2305.05176) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#explain-log-output

### 76. Explain a code snippet

> Explain what this function does, including its inputs, outputs, and side effects.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model when the complete function and relevant types are provided.
- **Why:** An explicit pattern plus executable checks gives a smaller model a bounded target and makes most mistakes visible.
- **Keep frontier when:** Architecture is changing, APIs are uncertain, security boundaries are involved, or the task requires long-horizon autonomous work.
- **Verify it:** Inspect the diff and run the formatter, typecheck, narrow tests, and relevant security checks.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#explain-code-snippet

### 77. Explain a formula

> Explain this spreadsheet formula in plain language with one worked example.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model, then test the worked example in the spreadsheet itself.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#explain-spreadsheet-formula

### 78. Arrange a fixed itinerary

> Arrange these confirmed stops into a two-day itinerary without adding places.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model when the locations, opening times, and travel constraints are already supplied and current.
- **Why:** The task is a bounded conversion from supplied operational material into a known structure or checklist.
- **Keep frontier when:** Priorities are disputed, missing context changes commitments, or the output authorizes an irreversible external action.
- **Verify it:** Trace each output item to the source, leave missing information unknown, and require a human approval before acting.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#plan-fixed-itinerary

### 79. Outline supplied material

> Turn this chapter into a hierarchical outline with section references.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model to mirror the supplied structure and cite the source section for every branch.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#outline-source-material

### 80. Build a dated timeline

> Put every dated event in these notes into chronological order.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Extract date-event pairs with source references, then sort them deterministically and flag partial dates.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#build-event-timeline

### 81. Describe an image

> Describe the main subject, setting, and visible action in this image.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient vision model for the first pass and preserve an uncertain state for ambiguous details.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#describe-image

### 82. Categorize images

> Sort these images into product, people, place, document, or other.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient vision classifier with a fixed label set and human review for multi-subject images.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#categorize-images

### 83. Extract text from a screenshot

> Extract the visible text from this screenshot and preserve the reading order.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use OCR or an efficient vision model, marking cropped and unreadable regions explicitly.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-screenshot-text

### 84. Caption an image collection

> Write one factual sentence describing each event photo.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient vision model for drafts, then verify people, location, and event context with the photo owner.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#caption-image-collection

### 85. Tag visible product types

> Tag the product types visibly present in each catalog photo.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient vision model for controlled product-type labels, not unsupported brand or model identification.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#tag-visible-products

### 86. Classify audio segments

> Label each segment as speech, music, silence, applause, or noise.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a specialized local audio classifier and review boundary timestamps.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#classify-audio-segments

### 87. Translate subtitles

> Translate these subtitles into French while preserving timestamps and line breaks.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with glossary constraints and deterministic subtitle-format validation.
- **Why:** The media input and requested metadata bound the output; the expensive step is usually review rather than frontier reasoning.
- **Keep frontier when:** The media is medically, legally, or technically specialized, or subtle context materially changes the answer.
- **Verify it:** Review against the original media, paying special attention to names, numbers, timing, and uncertain segments.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#translate-subtitles

### 88. Clean transcript formatting

> Add paragraphs and punctuation to this transcript without changing the words.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model with a verbatim constraint, then diff the token sequence after removing punctuation.
- **Why:** The supplied text contains the facts; the model is transforming expression rather than discovering new information.
- **Keep frontier when:** The copy creates legal commitments, handles a crisis or negotiation, or depends on a distinctive expert or creative voice.
- **Verify it:** Diff names, numbers, dates, links, claims, and promises against the source before publishing.
- **Evidence:** [Google](https://huggingface.co/google/gemma-4-E4B-it) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model)
- **Canonical entry:** https://youmightnotneedfrontier.com/#clean-transcript-formatting

### 89. Summarize a video transcript

> Summarize this video transcript with timestamps for each major point.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model grounded in timestamped chunks and verify each selected moment against the transcript.
- **Why:** The provided material bounds the answer, and explicit length, audience, and citation requirements constrain the transform.
- **Keep frontier when:** The work must reconcile conflicting sources, make a consequential judgment, or incorporate current facts beyond the supplied context.
- **Verify it:** Check every decision-relevant claim against the source and require section, page, line, or timestamp references where possible.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#summarize-video-transcript

### 90. Draft metadata from a transcript

> Draft a title, chapter list, and topic tags from this transcript.

- **Recommended route:** Local / open
- **Model policy:** Self-hosted. Gemma 4 leads this self-hosted shortlist. Hardware fit and task-specific evals still decide what works.
- **Three models to try:** [Gemma 4 E4B](https://huggingface.co/google/gemma-4-E4B-it) (Google) · [gpt-oss-20b](https://openai.com/open-models/) (OpenAI) · [Llama 4 Scout](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) (Meta)
- **Recommendation:** Use a local model for candidates, preserving transcript timestamps and separating generated titles from source facts.
- **Why:** The source and target fields bound the work, while a schema and source references make failures observable.
- **Keep frontier when:** Inputs are degraded, layouts are unusually complex, or a field requires legal, medical, financial, or domain judgment.
- **Verify it:** Validate the schema, retain source locations, and compare a representative sample with the original material.
- **Evidence:** [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) · [Simon Willison](https://simonwillison.net/2025/Feb/28/llm-schemas/) · [Stanford University](https://arxiv.org/abs/2305.05176)
- **Canonical entry:** https://youmightnotneedfrontier.com/#extract-video-metadata

### 91. Write tests for known behavior

> Add unit tests for these documented inputs, outputs, and edge cases.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model when expected behavior is explicit and the test runner is available.
- **Why:** An explicit pattern plus executable checks gives a smaller model a bounded target and makes most mistakes visible.
- **Keep frontier when:** Architecture is changing, APIs are uncertain, security boundaries are involved, or the task requires long-horizon autonomous work.
- **Verify it:** Inspect the diff and run the formatter, typecheck, narrow tests, and relevant security checks.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#write-unit-tests

### 92. Translate a small function

> Translate this Python function to TypeScript while preserving its tests.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model for a bounded function with existing tests and explicit runtime constraints.
- **Why:** An explicit pattern plus executable checks gives a smaller model a bounded target and makes most mistakes visible.
- **Keep frontier when:** Architecture is changing, APIs are uncertain, security boundaries are involved, or the task requires long-horizon autonomous work.
- **Verify it:** Inspect the diff and run the formatter, typecheck, narrow tests, and relevant security checks.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#translate-small-function

### 93. Build and test a regular expression

> Write a regex for these valid examples and reject these invalid examples.

- **Recommended route:** Use a tool
- **Model policy:** Tool-first. No general-purpose model is required. These Direct-class options can help configure, transform, or check the deterministic workflow.
- **Three models to try:** [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna) (OpenAI) · [Claude Haiku 4.5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.5 Flash-Lite](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite) (Google)
- **Recommendation:** Use a model to propose the expression if helpful, but let the supplied positive and negative fixtures decide.
- **Why:** The operation has a deterministic definition, so ordinary software can be exact, fast, and repeatable.
- **Keep frontier when:** The real problem is choosing ambiguous rules, resolving conflicting definitions, or interpreting consequential results.
- **Verify it:** Run the operation twice, preserve the source, and check counts or invariants before using the output.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Hamel Husain and Shreya Shankar](https://hamel.dev/blog/posts/evals-faq/)
- **Canonical entry:** https://youmightnotneedfrontier.com/#build-regular-expression

### 94. Explain lint errors

> Explain these lint errors and make the smallest fixes in the named file.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with the file, lint output, and exact lint command; reject unrelated edits.
- **Why:** An explicit pattern plus executable checks gives a smaller model a bounded target and makes most mistakes visible.
- **Keep frontier when:** Architecture is changing, APIs are uncertain, security boundaries are involved, or the task requires long-horizon autonomous work.
- **Verify it:** Inspect the diff and run the formatter, typecheck, narrow tests, and relevant security checks.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#explain-lint-errors

### 95. Write a query from a known schema

> Write a read-only SQL query for monthly active accounts using this schema.

- **Recommended route:** Efficient hosted
- **Model policy:** Direct → Standard. Standard class gives one class of cushion above this route’s Direct baseline.
- **Three models to try:** [GPT-5.6 Terra](https://developers.openai.com/api/docs/models/gpt-5.6-terra) (OpenAI) · [Claude Sonnet 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.6 Flash](https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash) (Google)
- **Recommendation:** Use an efficient model with the complete schema and definition, then inspect the query plan and sample output.
- **Why:** An explicit pattern plus executable checks gives a smaller model a bounded target and makes most mistakes visible.
- **Keep frontier when:** Architecture is changing, APIs are uncertain, security boundaries are involved, or the task requires long-horizon autonomous work.
- **Verify it:** Inspect the diff and run the formatter, typecheck, narrow tests, and relevant security checks.
- **Evidence:** [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#write-sql-from-schema

### 96. Plan a bounded schema migration

> Plan this nullable-to-required column migration with rollout and rollback steps.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Start balanced for the plan, insisting on measured data shape, compatibility phases, and a tested rollback.
- **Why:** The task still benefits from strong reasoning, but a tight scope and executable verifier can make a balanced model the economical first pass.
- **Keep frontier when:** Root cause is ambiguous across systems, rollback is weak, production data is at risk, or repeated checks keep failing.
- **Verify it:** Review the plan and diff, run the narrowest useful checks, and escalate on failed evidence instead of allowing improvisation.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Stanford University](https://arxiv.org/abs/2305.05176) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#plan-schema-migration

### 97. Debug a scoped regression

> Find why this tested endpoint began returning 500 after these three commits.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Start balanced when the regression window, logs, and reproduction are bounded; escalate if evidence crosses systems.
- **Why:** The task still benefits from strong reasoning, but a tight scope and executable verifier can make a balanced model the economical first pass.
- **Keep frontier when:** Root cause is ambiguous across systems, rollback is weak, production data is at risk, or repeated checks keep failing.
- **Verify it:** Review the plan and diff, run the narrowest useful checks, and escalate on failed evidence instead of allowing improvisation.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Stanford University](https://arxiv.org/abs/2305.05176) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#debug-scoped-regression

### 98. Triage a CI failure

> Triage this CI failure, identify the first causal error, and propose the smallest fix.

- **Recommended route:** Balanced hosted
- **Model policy:** Standard → Deep. Deep class gives one class of cushion above this route’s Standard baseline.
- **Three models to try:** [GPT-5.6 Sol](https://developers.openai.com/api/docs/models/gpt-5.6-sol) (OpenAI) · [Claude Opus 5](https://platform.claude.com/docs/en/about-claude/models/overview) (Anthropic) · [Gemini 3.1 Pro](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview) (Google)
- **Recommendation:** Use a balanced model with the full log, workflow, and reproducible check; require a passing rerun before acceptance.
- **Why:** The task still benefits from strong reasoning, but a tight scope and executable verifier can make a balanced model the economical first pass.
- **Keep frontier when:** Root cause is ambiguous across systems, rollback is weak, production data is at risk, or repeated checks keep failing.
- **Verify it:** Review the plan and diff, run the narrowest useful checks, and escalate on failed evidence instead of allowing improvisation.
- **Evidence:** [OpenAI](https://developers.openai.com/api/docs/guides/latest-model) · [Anthropic](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) · [Stanford University](https://arxiv.org/abs/2305.05176) · [SamRouter](https://router.samcarlton.com/about)
- **Canonical entry:** https://youmightnotneedfrontier.com/#triage-ci-failure

## Method

1. **Demand:** Assess scope, ambiguity, reasoning depth, tools, context, and verification burden.
2. **Readiness:** Check whether the prompt states the outcome, constraints, source context, success criteria, and non-goals.
3. **Risk:** Consider stakes, reversibility, privacy, approvals, and verifier strength.
4. **Route:** Try the cheapest plausible pass, check it, and escalate. Measured outcomes outrank a static leaderboard.

Search is lexical first. For queries of three or more characters, a pinned Supabase/gte-small ONNX embedding model runs locally in a Web Worker and appends meaning-based matches. The query stays in the browser.

## Sources

- **OpenAI: Using GPT-5.6.** Separates Sol, Terra, and Luna workload roles; recommends choosing effort intentionally and validating on representative work. https://developers.openai.com/api/docs/guides/latest-model
- **Anthropic: Choosing the right model.** Recommends an efficiency-first start for high-volume, straightforward work, with task-specific evals before upgrading. https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
- **Google: Gemini 3.5 Flash-Lite.** Positions the stable Flash-Lite model for high-volume, cost-sensitive classification, extraction, translation, and related workloads. https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite
- **Google: Gemma 4 E4B instruction-tuned model.** Documents on-device text, image, and short-audio input, including OCR and speech recognition, with a 30-second audio limit. https://huggingface.co/google/gemma-4-E4B-it
- **Simon Willison: Structured data extraction from unstructured content using LLM schemas.** Shows schema-constrained extraction across hosted and local models; useful as a concrete practitioner pattern, not a universal quality claim. https://simonwillison.net/2025/Feb/28/llm-schemas/
- **Hamel Husain and Shreya Shankar: LLM Evals FAQ.** Centers real prompts, human labels, held-out tests, and scoped checks when deciding whether a cheaper model is good enough. https://hamel.dev/blog/posts/evals-faq/
- **Stanford University: FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance.** Establishes the cheap-first cascade pattern: try a cheaper route, verify, and escalate when needed. https://arxiv.org/abs/2305.05176
- **SamRouter: How SamRouter classifies model demand.** Separates task demand, prompt readiness, execution risk, model class, and reasoning effort; the verifier remains part of the route. https://router.samcarlton.com/about

## Scope and cautions

- The examples are curated and intentionally not presented as a usage-frequency ranking.
- Provider guidance and model behavior change; recheck route logic before material use.
- High-stakes, irreversible, adversarial, ambiguous, or weakly verifiable work can justify frontier capability.
- This site does not collect prompts and does not claim lower-tier model equivalence or measured token savings.