Cursor's models fall into two groups: its own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5), which every paid plan covers from a weekly usage pool, and third-party models from Anthropic, OpenAI, Google and others, which you can use on Premium and higher at API list price. You pick one from the model picker in the chat panel, or let Auto choose for you.
This guide covers what is in the list as of October 2026, how Auto and Max Mode work, which plan unlocks what, and how to choose. Everything here comes from Cursor's Models & Pricing docs, the available models help page and cursor.com/pricing, read on 26 October 2026. The model list changes often, so check those pages before you rely on a specific name.
Which models are available in Cursor?
Cursor models (draw from your plan's weekly pool):
- Grok 4.7, Cursor's flagship, "jointly trained by Cursor and SpaceXAI." It has a 256k standard context window, a 500k long-context tier and four effort levels: low, medium, high and xhigh.
- Grok 4.6 and Grok 4.5, earlier models in the same family.
- Composer 2.5, Cursor's own agentic model, tuned for tool use, file edits and terminal work inside Cursor.
Most of these also come in a Fast variant that costs more per token.
Third-party models (not in the pool; Premium and higher, billed at API list price):
| Provider | Examples in the table |
|---|---|
| Anthropic | Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, Claude Fable 5.1 |
| OpenAI | GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5 |
| Gemini 3.1 Pro, Gemini 3.8 Flash, Gemini 4 Argon | |
| Others | Kimi K3 (Moonshot), GLM 5.3 (Z.ai), Muse Wonder (Meta) |
The full table lists many older models too, and a lot of them are marked "Hidden by default," so they will not show in the picker until you turn them on. Some notes in the table matter when you choose: several Claude models support "Up to 1M tokens with extended context at the same per-token rates," while the GPT-5.6 models support long context up to 1M tokens "with 2x input pricing."
How do you switch models in Cursor?
Cursor's prompting docs and keyboard shortcuts help page give three ways:
- Use the model picker dropdown at the top of the chat input.
- Press Cmd + / on Mac or Ctrl + / on Windows and Linux to cycle through models.
- Set a default model in Cursor Settings > Models.
The help center says your selection "persists across conversations until you change it." You can also switch in the middle of a chat, for example when a fast model did the exploring and you want a stronger one for the implementation. The active model always shows in the picker at the top of the chat panel. For more shortcuts, see the Cursor keyboard shortcuts guide.
How does Auto work in Cursor?
Auto means Cursor picks the model for each request. The help center describes it as selecting "models that balance intelligence, cost, and reliability," and recommends it for everyday tasks.
Behind Auto sits Cursor Router, which has three modes, picked under Optimize For in the model picker:
| Mode | What it does | Who gets it | Billing |
|---|---|---|---|
| Cost | The previous Auto logic, optimized for token spend | Premium, Super, Ultra, Teams, Enterprise | A Cursor model request from the weekly pool |
| Balance | Balances intelligence, speed and cost | Teams and Enterprise with Router on | The routed model's list price |
| Intelligence | Sends harder tasks to the most capable models | Teams and Enterprise with Router on | The routed model's list price |
Two details trip people up:
- Plus does not include Auto. The plan list says Plus has "No Auto," along with no third-party models.
- Auto does not control subagents. The subagents docs say "The model picker, including Auto, applies to the parent agent." A subagent can run a named third-party model, and that usage bills as third-party usage even if your picker shows Auto.
By default the model that Router chose stays hidden, so "you judge results on merit." Team admins can switch this to Displayed. The Cursor Router docs cover the admin controls.
What is Max Mode in Cursor?
Max Mode extends a model's context window beyond its default. As of October 2026 it is a legacy feature: the Max Mode help page says it "is available only on legacy request-based plans," and the usage limits page says "Current plans don't include Max Mode."
If you are still on a legacy request-based plan:
- Toggle Max Mode in the model selector of the chat or agent panel. It stays on across conversations.
- Models that require it turn it on automatically when you select them. Many rows in the pricing table say "Requires Max Mode on legacy request-based plans."
- It is billed at the model's API rate plus 20%, and a bigger window can mean more tokens per request.
Cursor's own advice: "The default context window works well for most coding tasks." On current plans, long context is handled per model instead, for example Grok 4.7's 500k tier or the 1M-token options on Claude and GPT models.
Which plans get which Cursor models?
As of October 2026, per Models & Pricing:
| Plan | Price | Cursor models | Third-party models | Auto |
|---|---|---|---|---|
| Free | $0 | Limited Agent and Chat | Not available | Not stated |
| Plus | $8/mo | Yes, fixed settings | Not available | No |
| Premium | $30/mo | Yes, 4x Plus usage | API list price | Cost mode |
| Super | $100/mo | Yes, 15x Plus usage | API list price | Cost mode |
| Ultra | $200/mo | Yes, 50x Plus usage | API list price | Cost mode |
| Teams | $30, $100 or $200 per seat/mo | Matches seat tier | API list price plus Cursor Token Rate | All three modes with Router |
"Fixed settings" on Plus means, in Cursor's words, "Fast is off, effort is medium, and Grok 4.7 uses the 256k context window." Moving to Premium or higher lets you "choose Fast mode and effort levels." Legacy Pro, Pro+ and Ultra plans also unlock third-party models, from a separate monthly Other Models pool. The Cursor pricing guide covers the plans in more depth.
A few related rules:
- Your own API keys (Cursor Settings > Models) work on Premium and higher but not on Plus, and your provider bills the model cost directly. On Teams and Enterprise the Cursor Token Rate still applies to eligible models. The API keys page notes that "Tab completion continues using Cursor's built-in models."
- Regions: some models are unavailable in some regions because of provider restrictions. Cursor says Auto "is available in all regions."
How do you choose a Cursor model?
Start with Cursor's own suggestions from the available models page:
- Auto for everyday tasks.
- Grok 4.7 for "the hardest, longest-running tasks."
- Composer for fast, cost-efficient interactive coding. Cursor says "Composer is not being deprecated."
- Claude Opus and GPT-5.6 Sol "handle complex, multi-step tasks well."
- "Some users also prefer Gemini Pro models."
Then weigh cost. Cursor models come out of your weekly pool, while third-party models are paid per token from the first request. Per million tokens, as of October 2026, Composer 2.5 costs $0.5 input and $2.5 output, Grok 4.7 costs $2 and $6, and Claude Opus 5.5 costs $4 and $20. Fast variants cost more: Composer 2.5 Fast is $3 input and $15 output.
A practical pattern: use Composer or Auto for quick edits and exploration, and switch to Grok 4.7 at high effort or a frontier third-party model when a task spans many files or needs careful reasoning. Plan mode pairs well with a stronger model for the planning step.
Keep your prompts portable across models
Whichever model you pick, the prompts and Agent Skills you reuse are worth keeping outside any one tool. Kipt stores them as Markdown in a private GitHub repo you own, with {{name}} blanks, tags and Cmd+K search, so the same prompt works whether you send it to Composer, Claude or GPT. For writing better prompts in Cursor, see the Cursor prompts collection.