Choosing models

Each agent request is answered by a model. Cursor offers its own models and models from companies such as OpenAI, Anthropic and Google; which ones you can use depends on your plan, and names change often.

How to choose:

• Auto is the default for everyday work. Cursor's router picks a model per request, balancing capability, cost and reliability, and the chosen model can change between turns.
• The model picker in the chat panel lets you pick a specific model; Cmd+/ or Ctrl+/ cycles through models. A change applies from that point in the conversation.
• Set your default in Cursor Settings > Models. Your selection persists across conversations until you change it.
• You can add your own API key for a provider in Cursor Settings > Models, for example when a model isn't offered in your region. Requests still pass through Cursor's servers.

Matching model to task:

• Quick edits, small questions, exploring: a fast model is enough and keeps you in flow.
• Multi-file refactors, tricky bugs, architecture: a more capable reasoning model, which "thinks" step by step before answering, is worth the extra time and cost.
• A useful pattern: explore and plan with a fast model, then switch to a stronger one for implementation.

Models also differ in context size (how much text they can consider). Larger isn't automatically better: stuffing more into context costs more and can distract the model. When a model struggles repeatedly, a new chat with a clearer prompt often helps more than switching models.