What models does Cursor use?
Cursor currently separates chat and agent models into two pools. The first-party Cursor Models pool includes Cursor Grok 4.5 and Composer 2.5. Other Models covers third-party models from providers such as Anthropic, OpenAI and Google. Auto is a routing choice, not one fixed model: Cost, Balance and Intelligence can route requests differently. The catalog can change, so verify it in Cursor's model picker or official model guide. Verified 2026-08-09.
Cursor's model choices at a glance
| Choice | What it includes | Who selects the model | Usage treatment |
|---|---|---|---|
| Cursor Models | Cursor Grok 4.5 and Composer 2.5 | You select a listed Cursor model | Draws from the Cursor Models pool |
| Other Models | Third-party models from labs such as Anthropic, OpenAI and Google | You select a listed third-party model | Draws from the Other Models pool at the applicable model rate |
| Auto | Cost, Balance or Intelligence routing modes | Cursor Router selects per request | Pricing depends on the selected Auto mode and routed model |
Why — the first-principles explanation
The phrase first-party models is easy to misunderstand. Cursor's current documentation calls the category Cursor Models and lists two models in it: Cursor Grok 4.5, jointly trained by Cursor and SpaceXAI, and Composer 2.5, Cursor's own agentic coding model. That is a product and billing category; it does not mean every model visible in Cursor was trained only by Cursor.
The second category is Other Models. It contains third-party models from external labs. The exact catalog changes as models ship or retire, so a durable answer should name the providers and explain the category instead of freezing a long model list that will age immediately. When you manually select one of these models, usage comes from the Other Models pool under the current plan rules.
Auto is not another fixed first-party model. On supported plans, Cursor Router can choose a model for each request. Auto Cost uses bundled pricing; Auto Balance and Auto Intelligence route according to task complexity and are billed under their documented model-rate rules. You choose the optimization mode, while Cursor chooses the underlying model.
That gives you two separate decisions: choose a named model when repeatability or a specific capability matters, or choose Auto when you want Cursor to trade off cost, speed and intelligence per task. For billing questions, check the two usage pools separately rather than treating every model request as one allowance.
An example that makes it click
Think of the model picker as a train station. The Cursor Models platform currently has two named trains, Cursor Grok 4.5 and Composer 2.5. The Other Models platform hosts trains operated by outside providers. Auto is the dispatcher: you choose whether it should optimize for Cost, Balance or Intelligence, and it assigns a train for each trip. The platforms, dispatcher and fare rules are related, but they are not the same thing.
How to do it
- Open the model picker in Cursor's chat or agent panel.
- Choose a named Cursor Model when you want Composer 2.5 or Cursor Grok 4.5 specifically.
- Choose a named Other Model when you need a particular third-party model and accept its current usage rate.
- Choose Auto, then Cost, Balance or Intelligence, when you want Cursor Router to select per request.
- Open the Usage dashboard and check Cursor Models and Other Models separately before assuming which allowance a request used.
- Recheck the official model guide when a model name matters, because the available catalog and routing pool can change.
Key facts
- Cursor's current Cursor Models pool includes Cursor Grok 4.5 and Composer 2.5 (verified 2026-08-09).
- Composer 2.5 is Cursor's own agentic coding model; Cursor Grok 4.5 is jointly trained by Cursor and SpaceXAI.
- Other Models is the separate pool for third-party models; the exact catalog changes over time.
- Auto is routing, not one fixed model: Cost, Balance and Intelligence have different optimization and pricing behavior.
- With Auto Balance or Intelligence, Cursor Router chooses an underlying model for each request.
- The model picker and Usage dashboard are the authoritative account-level views when documentation and an older article differ.
Use the vendor's current model documentation to verify the model family, routing behavior and product terminology.
Understand Cursor models before changing billing controls
Use the model taxonomy page for model choice and the linked controls pages for allowances, spend caps and what happens at a limit.
▶ The 60-second explainer (script)
What models does Cursor use? Its current model picker separates Cursor Models from Other Models. The Cursor Models pool lists Cursor Grok 4.5, jointly trained by Cursor and SpaceXAI, and Composer 2.5, Cursor's own agentic coding model. Other Models contains third-party choices from providers such as Anthropic, OpenAI and Google. Auto is not a separate model: Cost, Balance and Intelligence are routing modes that can select an underlying model for each request. Pick a named model when repeatability matters, or use Auto when you want Cursor to trade off cost and capability. The catalog changes, so confirm the current names and usage pool in Cursor's official model guide and your account dashboard. Verified 2026-08-09.
What authoritative sources say
People also ask
What are Cursor's first-party models?
Cursor's current documentation calls the category Cursor Models and lists Cursor Grok 4.5 and Composer 2.5. The lineup can change, so confirm it in the official model guide.
Is Auto a first-party model?
No. Auto is a routing choice. Depending on the mode, Cursor Router selects an underlying model for each request.
Can I choose which model Cursor uses?
Yes. Pick a named model for direct control, or choose Auto and an optimization mode to let Cursor route the request.
Does the model I pick affect my usage?
Yes. Cursor Models and Other Models use separate pools, and Auto's treatment depends on the selected mode and routed model. Check the current Usage dashboard for your account.
The same question, asked other ways
- Which AI models does Cursor support?
- Does Cursor use Claude?