Cursor changes its model menu frequently. As of June 23, 2026, the official model documentation lists current families from Anthropic, OpenAI, and Google. Treat exact model availability, pricing, and provider routing as volatile.

The durable skill is choosing the right class of model for the work.

Start with Auto

Cursor’s Auto mode routes work across models based on capability, cost, and reliability. Use it for ordinary implementation, refactoring, and bug fixing until you have evidence that a specific model is better for the current task.

Auto is also a good default for teaching because it keeps the lesson focused on the workflow instead of model trivia.

Escalate for Hard Reasoning

Choose a stronger explicit model when the task has one of these traits:

• Ambiguous architecture decisions.
• Large cross-file refactors.
• Security-sensitive code.
• Complex migrations with many edge cases.
• Reviews where false confidence is expensive.

Ask for a plan first, then decide whether the plan justifies the slower or more expensive model.

Use Max Mode Deliberately

Max Mode gives Cursor access to the model’s larger context window and premium reasoning path. It is useful when the task genuinely depends on a broad slice of the repository. It is wasteful when the task needs three files and a test.

Cloud Agents always run with Max Mode enabled, so write cloud prompts with the same discipline you would use for an expensive local session: exact scope, acceptance criteria, and verification commands.

Keep Model Claims Dated

If you write a rule, tutorial, or team process that names a model, date the claim:

As of 2026-06-23, use Cursor Auto for default implementation work and escalate to
the strongest available reasoning model for architecture reviews.

That phrasing survives menu changes. “Always use Model X” usually does not.