Pick a model for your task

Match the job to model type before chasing the biggest name on the leaderboard.

| Task | Start here | Why |

|------|------------|-----|

| Everyday chat & drafts | General-purpose chat model (GPT-4o-mini, Claude Haiku, etc.) | Cheap, fast, good enough |

| Long PDFs / contracts | Long-context model | Fits more text in one pass |

| Code generation | Code-tuned or strong reasoning model | Better syntax & debugging |

| Images | Diffusion / image API | LLMs don't draw pixels |

| Embeddings for RAG | Dedicated embedding model | Cheaper + better retrieval |

| Local/offline | Small open-weight + quantization | Privacy, no API bill |

1. **Can prompt + docs solve it?** → cheapest chat model + RAG if needed

2. **Need tools/actions?** → model with reliable function calling

3. **Need lowest latency?** → smaller model or streaming

4. **Need highest quality on hard evals?** → flagship model for hard cases only

Route easy queries to small models; escalate to large models only when confidence is low or user asks for "best quality."

**Related lessons:** `choosing-a-model`, `pick-the-right-tool`