2 min lesson
Two groups of models, two jobs
Imagine this comes up at work. "A developer's agent keeps calling a method that doesn't exist in their library, with full confidence. Which diagnosis and workaround fits?" Start with the practical move.
Step 1 of 3
Two groups of models, two jobsmatch the model to the task
Sort models by what they're for before asking which is best: an async, high-reasoning group for planning and auditing, and a fast, synchronous group for implementing.
Strong reasoning, excellent at long-horizon tasks, more costly per token.
Reach for planning, plan mode and auditing. The GPT-5.x line and Claude Opus, today Opus 5.5, sit here.
Near-frontier intelligence at much higher speed: the implementers.
Cursor's in-house 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. lives here, fast and reliable at medium-difficulty work, with frontier fast variants for documentation. Cursor's own pool also covers Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition., 4.6 and 4.5, so check each model's speed and intelligence ratings in the docs before sorting it.
Interactive widget. Tab through its controls; the result updates in the panel below as you change them.
Sort by job first, reasoning for planning and a fast executor for implementation, then let Auto route the choice on plans that include it. Balance and Intelligence routing is Teams and Enterprise only.