Which Model and Context Size Should You Pick in Cursor?
Rule of thumb, not a vendor recommendation: on The X Subscription, start with Cursor's own models, since those draw on your plan's weekly pool: Composer 2.5 for interactive coding and Grok 4.7 for the hardest long-running tasks. Reach for Claude, GPT or Gemini when you've got a reason, because they're paid from prepaid credit. Press Cmd+/ (Ctrl+/) to cycle through models, or use the picker in the input.
Model default: Cursor Settings > Models Cycle models: Cmd+/ (Ctrl+/) Auto mode: open the model picker, select Auto, then choose a mode To compare on a real task, with a worktree for each run: /best-of-n composer,sonnet,gpt [the same task] Then ask: Compare these results on correctness, how closely they follow @[reference file], and diff size. Recommend one, and say what would change your mind.
- 1Composer 2.5interactive work, weekly pool
- 2Grok 4.7long, hard tasks, weekly pool
- 3Claude, GPT, GeminiPremium and higher, prepaid credit
Try it now
- 1Open **Cursor Settings > Models** and check which models you can see on your plan.
- 2In a chat, press Cmd+/ (Ctrl+/) until you reach a different model. The active one shows in the model picker.
- 3Run the same small task on two models, and compare the diffs. If you can't point to a difference that matters, keep the one that costs less.
Use it when
When a task gets harder or more routine than the last one. Cursor's docs note you can use a faster model for exploration and switch to a stronger one for implementation in the same conversation, and the change applies going forward.
Skip it when
Don't spend time comparing models on trivial edits, where the difference is rarely visible. And don't switch models hoping to fix a context problem. If Agent is editing the wrong file, a costlier model will usually edit the wrong file more expensively.
Why it works
The model affects both results and what pays for them. Cursor's docs price Composer 2.5 at $0.50 per million input tokens and $2.50 output, Grok 4.7 at $2 and $6, and Claude Sonnet 5.5 at $2 and $10. On The X Subscription, third-party models bill at list price from request one. Auto picks for you where it's available. Cost mode, or classic Auto, draws from the weekly pool whichever model serves it. Balance and Intelligence are for Teams and Enterprise with Cursor Router on, and bill at the routed model's list price.
Example
Say you're on Premium. You use Composer 2.5 for a day of routine edits. For a large concurrency bug, you switch to Grok 4.7 in the same chat. Your subscription covers both from the weekly pool. For a security-sensitive review, you try Claude Opus 5.5, knowing it's $4 per million input tokens from prepaid credit. Rough arithmetic, not a vendor figure: one request with 100,000 input and 5,000 output tokens is about $0.50 before cache discounts.
Going further
Grok 4.7 long context, above 256k input tokens, is billed at 2x the standard rate, up to a 500k window, so choose the larger window only for tasks that need it. Plus is fixed to the 256k window, medium effort, with Fast off. Fast variants cost more too. Max Mode now exists only on legacy request-based plans. Watch subagents, which can run a named third-party model while your picker shows Auto or Composer. Legacy plans differ, since Pro keeps a monthly allowance for other models, so check which system a guide refers to.
Common mistakes
- Assuming Claude is included in your plan. On The X Subscription, it isn't.
- Choosing a 500k window by default and paying for space you aren't using.
- Following a tutorial about Pro's limits when you're on a new plan, or the reverse.
- Expecting to see which model Auto used. On teams, the routed model stays hidden unless an admin sets it to Displayed.
Sources
- Models & Pricing, Cursor Docs
- Available models, Cursor Help
- Cursor Router, Cursor Help
- Usage and limits, Cursor Help
- Prompting agents, Cursor Docs
Checked 6 November 2026. Products change often, so confirm details with the vendor.