2 min lesson
Composer and Auto
Put yourself in this case: "A platform lead worried about runaway spend asks how their engineers should choose models. What's the strongest, most concrete answer?" Give the clearest next step.
Step 1 of 2
Composer and Autothe daily driver and the router
ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. is Cursor's own model, trained specifically for writing code - more focused and lower resource consumption than general-purpose LLMs, and extremely fast, which keeps developers in flow. Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. launched as near-frontier intelligence at roughly a tenth of the cost; it's notably better at judging high- versus low-effort tasks so it doesn't over-engineer. It's the recommended default for execution.
Auto mode runs a router that reads prompt complexity and routes under the hood, defaulting to ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. and factoring in model availability. The non-obvious detail worth coaching: Auto is steered by qualitative cues - words like "quickly" or "more accurately" - so engineers should be explicit about the characteristics they want. You can't see the exact per-step model breakdown in chat; infer it from the console usage page filtered to your user. New users are opted into Auto by default.
When a platform lead asks "which model," they're really asking how to keep spend predictable while output stays good. Answer with the loop (plan with a frontier model, execute with ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition.), the brain-icon cue so engineers self-select reasoning power, and Auto as the sane default for people who don't want to think about it. That reframes the question from a taste debate into a governable habit - which is exactly the answer that lands with someone responsible for the bill.
Max Mode's bigger context window is genuinely useful for auditing a sprawling codebase, but a larger context degrades accuracy and costs more. If you pitch it as a free upgrade, a sharp engineer will catch the tradeoff and discount you. Frame it as a selective tool used in plan mode on tasks that actually need the extra short-term memory.
Learn more
Full explanation
Full explanation
When the model question comes, don't say "it depends." Name the three categories, the brain-icon cue, the plan-expensive/implement-cheap loop and what Auto routes to. Then tie it to the customer's concern - predictable spend with good output. Depth on model selection is one of the clearest signals that you actually use the product, because it's the question every platform team raises and most candidates fumble.
Learn more
Optional practice