2 min lesson
Model selection: the question platform teams ask first
Compare two rows from "Model selection: the question platform teams ask first", then say when each one fits.
Step 1 of 2
A platform team will ask which model their engineers should use on day one, and "it depends" is the answer that loses the room. The reciting-features version of this question is everywhere; the depth that marks a real operator is knowing the two model types, the cost loop and what Auto does on each plan.
Two model categories matter in daily use. A standard model thinks once and responds, which makes it the fastest and cheapest. A thinking (reasoning) model runs chain-of-thought, so it is slower and consumes far more tokens; reserve it for deep reasoning, planning or when a standard model has hit a wall. Max Mode exists only on legacy request-based plans, where it extends the context window and bills at the model's API rate plus 20%. On current plans, larger context is a per-model Max Context option, and a bigger window still costs accuracy and tokens.
- Category
- Standard
- How it behaves
- Thinks once and responds
- When it's the right call
- Fast, cheap, everyday completions and small changes
- Category
- Thinking / reasoning
- How it behaves
- Chain-of-thought, far more tokens
- When it's the right call
- Planning, large refactors, when a standard model is stuck
| Category | How it behaves | When it's the right call |
|---|---|---|
| Standard | Thinks once and responds | Fast, cheap, everyday completions and small changes |
| Thinking / reasoning | Chain-of-thought, far more tokens | Planning, large refactors, when a standard model is stuck |
Two live categories. Learn the split, not the roster, because the named models turn over every few months.
A complementary split runs along the roster. Frontier and high-reasoning models such as Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., GPT-5.6 Sol and Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. suit large refactors and detailed planning. Daily-driver models such as 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. and the Cursor Grok 4.x models suit easy implementation and small changes.