AgentdexBlog · · Agentdex Editorial
What AI Does Cursor Use? (2026 Answer)
Cursor (Agentdex profile) uses a mix of its own models and models from other AI labs. Its own lineup is the Grok family (Grok 4.7, 4.6 and 4.5) plus Composer 2.5, and on higher plans you can also pick third-party models from Anthropic, OpenAI and Google. Which AI actually answers your prompt depends on your plan, the feature you are using and whether you choose a model yourself or leave it on Auto. This guide explains each piece, based on Cursor's models documentation and announcements as of October 24, 2026. For plan prices, see our Cursor pricing guide.
Key Takeaways
- Cursor's own models are Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5. Cursor's docs list Cursor itself as the provider for the Grok models, which it has trained jointly with SpaceX's AI team since SpaceX bought Cursor in 2026.
- Composer 2.5 is built on Kimi K2.5, an open-weight model from Moonshot AI, with further training by Cursor. It is the cheapest model Cursor offers.
- Third-party models are available on Premium and above. These include Claude (Sonnet 5.5, Opus 5.5, Fable 5.1, Haiku 5.5), Gemini (4 Argon, 3.1 Pro, 3.8 Flash) and GPT-5.6 (Sol, Terra, Luna). They are billed at API list price, outside your plan's included usage.
- The Plus plan only covers Cursor's own models, at fixed settings, with no Auto mode and no option to use your own API keys.
- OpenAI models are on a countdown. OpenAI plans to stop supplying models to Cursor, with a proposed cutoff of November 12, 2026, and will not provide new models.
The short answer
There is no single "Cursor AI". Cursor is an editor and agent that routes your requests to a language model. The model can be one Cursor trains and serves itself, or one from another lab that Cursor connects to. Cursor's Models & Pricing documentation says it supports frontier models from OpenAI, Anthropic, Google and SpaceX's AI lab, and splits them into two groups: Cursor models, which are covered by your plan's weekly usage pool, and third-party models, which are not.
| Group | Models (October 2026) | How it is paid for |
|---|---|---|
| Cursor models | Grok 4.7, Grok 4.6, Grok 4.5, Composer 2.5 | Included in every paid plan's weekly usage pool |
| Anthropic | Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1, Claude Haiku 5.5 | API list price, from prepaid credit (Premium and above) |
| Gemini 4 Argon, Gemini 3.1 Pro, Gemini 3.8 Flash | API list price, from prepaid credit (Premium and above) | |
| OpenAI | GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna | API list price, from prepaid credit (Premium and above), until the OpenAI cutoff |
The list changes often. Treat the table as a snapshot and check the model picker in your editor for what you can use today.
Cursor's own models: Grok and Composer
Grok 4.7, 4.6 and 4.5
Since Cursor joined SpaceX (the deal closed on August 14, 2026), the Grok models have become Cursor's default family. Cursor's docs list Cursor as the provider for Grok 4.7, which has a 256k-token context window by default and up to 500k on the long-context variant. Grok 4.7 is the most capable of the three and the one most people should use for agent work. Grok 4.6 also matters behind the scenes: Cursor says it must be enabled for Cursor Router to work on team plans.
Each Grok model has a standard and a "Fast" variant. Fast responds quicker but costs more when you are paying for extra usage. On the Plus plan, Fast is off and reasoning effort is fixed at medium.
Composer 2.5
Composer is Cursor's in-house coding model line. Composer 2.5 is built on Kimi K2.5, an open-weight model from Moonshot AI, which Cursor then trained further for agentic coding inside its editor. It is aimed at fast, low-cost agent work, and its extra-usage rates are much lower than Grok's: Cursor lists $0.50 per million input tokens and $2.50 per million output tokens for standard Composer 2.5, against $2 and $6 for Grok 4.7.
That makes Composer 2.5 a sensible choice for routine edits, refactors and quick questions, while Grok 4.7 or a frontier third-party model handles harder planning and debugging.
Third-party models: Claude, Gemini and GPT
On Premium ($30/month) and higher, you can choose models from other labs. Cursor does not include them in your weekly pool. You pay the provider's API list price from prepaid credit, and on Teams and Enterprise there is also a Cursor Token Rate on eligible models. In practice that means a heavy Claude Opus user should budget for the plan price plus model usage.
- Anthropic: Claude Sonnet 5.5 for everyday agent work, Opus 5.5 and Fable 5.1 for long, complex tasks, and Haiku 5.5 for fast, cheap answers.
- Google: Gemini 4 Argon for complex coding and reasoning, Gemini 3.1 Pro, and Gemini 3.8 Flash for speed.
- OpenAI: GPT-5.6 Sol, Terra and Luna, which range from deep reasoning (Sol) to quick, cheap tasks (Luna).
What happens to OpenAI models in Cursor
After SpaceX acquired Cursor, OpenAI said it would end its model supply. According to OpenAI's help article, the proposed cutoff is November 12, 2026, though that date could change. OpenAI will not provide future models to Cursor. After the cutoff, you can still use OpenAI models in local Chat and Agent with your own OpenAI API key, but not in Tab, Auto, cloud agents, Automations, the CLI or the Cursor API and SDK.
If your workflow depends on GPT models, plan the switch now: try Claude Sonnet 5.5 or Grok 4.7 on the same kinds of tasks before the cutoff, so you are not changing models in the middle of a deadline.
Which model Cursor uses for each feature
| Feature | What powers it |
|---|---|
| Agent and Chat | The model you pick in the model picker, or Auto |
| Auto (individual plans) | Cursor's routing picks a model for you; not available on Plus |
| Auto (Teams and Enterprise) | Cursor Router, with Cost, Balance or Intelligence modes |
| Tab completions | Cursor's own completion model, not one you choose |
| Cloud agents and Automations | Your chosen model or Auto (Premium and above) |
| Bugbot code review | Managed by Cursor (Premium and above) |
Auto and Cursor Router
Auto lets Cursor choose the model. On Teams and Enterprise plans, Auto is powered by Cursor Router, which runs a classifier on each agent request and sends it to the most cost-effective model that should still give comparable quality. You steer it with one of three modes:
- Cost uses Cursor's earlier Auto logic and optimizes for token spend.
- Balance trades off cost and quality.
- Intelligence leans toward the most capable models.
You cannot hand-pick which model handles a routed request, and the pool changes as new models ship. When the router sends a request to a third-party model, it bills as third-party usage. If your team blocks too many models, routing quality drops and the router can stop working.
Tab completions
Tab, Cursor's multi-line autocomplete, runs on Cursor's own completion model. You do not select it in the model picker, and it does not depend on which chat model you choose. That is also why the OpenAI cutoff and bring-your-own-key rules do not change how Tab works.
What you get on each plan
| Plan | Cursor models | Third-party models | Auto |
|---|---|---|---|
| Free ($0) | Limited Agent and Chat | No | Not listed |
| Plus ($8/month) | Yes, fixed settings (Fast off, medium effort) | No, and no own API keys | No |
| Premium ($30/month) | Yes, 4x Plus usage | Yes, at API list price | Yes |
| Super ($100/month) | Yes, 15x Plus usage | Yes, at API list price | Yes |
| Ultra ($200/month) | Yes, 50x Plus usage | Yes, at API list price | Yes |
| Teams and Enterprise | By seat tier or pooled | API list price plus Cursor Token Rate on eligible models | Cursor Router |
Prices are from Cursor's docs as of October 24, 2026. Older guides that mention Pro ($20) or Pro+ describe legacy plans; existing subscribers on those plans keep them, but new buyers see the plans above.
Can you use your own API key?
Partly. On Premium and above, you can add your own provider keys for Chat and Agent. Plus does not allow it. Own keys do not power Tab, and, as OpenAI's help article notes for OpenAI keys after the cutoff, they do not work with Auto, cloud agents or Automations. If you want full control over which model runs and who you pay, an open-source agent such as Cline (Agentdex profile), which works inside VS Code or Cursor, is the more flexible route. Our Cline vs Cursor comparison covers the trade-offs.
How to choose a model in Cursor
- Default: Grok 4.7 for agent work if you are on Plus, or Auto if you are on Premium and want Cursor to manage cost.
- Routine edits and quick questions: Composer 2.5. It is fast and the cheapest option when you go beyond your pool.
- Hard bugs and big refactors: Claude Opus 5.5, Gemini 4 Argon or GPT-5.6 Sol on Premium, accepting that you pay API rates.
- Long sessions on a budget: stay on Cursor models so usage comes from your weekly pool, and switch to a third-party model only for the steps that need it.
Model quality varies by language, codebase and prompt, so the best test is your own: run the same small task through two models and compare the diffs. Clear rules files help any model; our Cursor rules guide shows how to set them up.
Privacy: where your code goes
When you use a model, the relevant code and prompt are sent to whoever serves it: Cursor for its own models, or the third-party provider through Cursor. Cursor's docs mark which models offer US data residency, and Cursor offers a Privacy Mode and team privacy controls. If you work on sensitive code, read Cursor's security page and your company's policy before choosing a third-party model.
FAQ
Does Cursor use ChatGPT?
Not ChatGPT itself. Cursor has offered OpenAI's GPT models, currently GPT-5.6 Sol, Terra and Luna, on Premium and above. OpenAI plans to stop supplying them, with a proposed cutoff of November 12, 2026. After that, you can only use OpenAI models in local Chat and Agent with your own OpenAI API key.
Does Cursor use Claude?
Yes, on Premium and higher plans. Claude Sonnet 5.5, Opus 5.5, Fable 5.1 and Haiku 5.5 are available and billed at Anthropic's API list price, outside the plan's included usage. They are not available on Plus.
Is Grok the default model in Cursor now?
Grok models are Cursor's own models and the ones every paid plan's weekly usage covers. On Plus they are the only option, alongside Composer 2.5. On higher plans you can still choose Claude, Gemini or GPT, or let Auto decide.
What is Composer in Cursor?
Composer is Cursor's in-house coding model line. Composer 2.5 is built on Moonshot AI's open-weight Kimi K2.5 and further trained by Cursor for agentic coding. It is the lowest-cost model Cursor offers for extra usage.
Which AI model does Cursor Tab use?
Tab uses Cursor's own completion model. You cannot swap it for a chat model, and your own API keys do not apply to it.
Can I use local models in Cursor?
Cursor is built around cloud models and does not offer a built-in local model option. If running models locally matters to you, look at open-source agents such as Cline or Kilo Code with Ollama or LM Studio.
Written by Agentdex Editorial. Agentdex is an independent directory reviewing AI coding tools; we are not affiliated with the companies mentioned. Prices and features are as of October 24, 2026, from official sources linked above. Always confirm on the vendor's site.