Cursor Models: Which Model Should You Use?

Cursor groups its current models into two usage pools. Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. contains Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5 and Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition.. Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. contains the third-party options from Anthropic, Google and OpenAI. The split matters because each pool has its own included usage, and the Start plan in India includes only Cursor Models.

Gemini 3.7 FlashGoogle's latest speed-tier model in Cursor for high-throughput coding, with a 1M maximum context window and a 90% cached-input discount. Press Enter for the full definition. is Google's current speed-tier choice in that second pool. It succeeds Gemini 3.6 Flash with lower published rates, while Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. remains the more specific starting point for visual coding and Pro-tier reasoning. Keep the 3.6 page only for a controlled predecessor comparison or an older policy reference.

Do not confuse a model with the product that runs it. Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. is a persistent teammate with apps, routines, a shared cloud computer and separate eligibility rules. Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. is a model you select for a request. The overlap in names does not make their setup, data boundaries or billing interchangeable.

Use the least expensive model that can finish the task cleanly. That is not the same as using the cheapest model for every first attempt. A weak attempt creates review work, fills the conversation with a bad path and often forces a second run. The table below is a starting policy, not a benchmark result.

Model names are a poor substitute for a good task boundary. Give the model the relevant files, the expected behavior and one check it can run. Then judge the diff and the evidence, not the confidence of the answer. A model that finishes in one pass on your repository is often cheaper than a lower-rate model that needs three retries.

The two pools reset with the monthly billing cycle. Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5 and Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. draw from Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition.. A model you select from Anthropic, Google or OpenAI draws from Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. at its listed API price. Different token rates mean two models can consume the same pool at very different speeds.

Teams and Enterprise accounts also need to account for the Cursor Token Rate on eligible third-party work and any data-residency uplift in their agreement. The model's headline input and output rates are therefore a starting point, not always the whole team cost. Keep the worked arithmetic in your rollout document, beside the model allow-list that makes the choice possible.

On Teams and Enterprise, Auto uses Cursor Router. You choose Cost, Balance or Intelligence, and Cursor chooses an underlying model for each request. Cost keeps the bundled Auto pricing path. Balance and Intelligence bill at the rate of the model selected for the request, so they can consume usage faster than Cost.

• Use Cost when predictable bundled usage matters most.
• Use Balance when a team wants one practical default across mixed daily work.
• Use Intelligence for a task mix that justifies stronger models and the higher average spend.
• Pin a model when you need a reproducible comparison, a specific privacy approval or a known provider capability.

Router results depend on the models an admin allows. Cursor warns that blocking too many models reduces routing quality and can disable routing. Treat the model allow-list and the Router mode as one decision. If a team changes either one, rerun the same small evaluation set and inspect cost, acceptance and review time again.

No. Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. means Cursor and its model providers do not train on your code data, but model retention and regional processing still vary by model and route. Cursor's privacy documentation names Fable 5 as a special case that requires an explicit opt-in to 30-day provider retention, so it does not fit a Zero Data Retention requirement.

Do not infer a model's approval from its family name. Check the current privacy table for the exact model, the region you selected and whether you use Cursor routing or your own API key. For a team rollout, record that review next to the model allow-list and repeat it when a new model replaces the old default.

Cursor StartAn India-only individual plan centered on Cursor Models, with cloud and mobile access but without Other Models, Auto, Bugbot, Automations or the Agent SDK. Press Enter for the full definition. includes Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5 and Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. through the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool. All three run in non-fast mode. Both Grok models use fixed medium effort, so Start users cannot change effort or enable Fast. The plan does not include Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition., Auto, on-demand usage, Automations, BugbotCursor's automated PR reviewer that posts inline findings and can push fix commits from isolated VMs. Press Enter for the full definition. or the Cursor SDK.

That makes the choice simpler. Use Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. for ordinary work and move to Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. when the job needs more depth. Upgrade only when the missing capability has a real job attached to it, such as a third-party model required by policy, Router for a team, or on-demand usage after the included pool is gone.

Run a small evaluation with real work, not a general coding benchmark. Choose three to five tasks your team already understands: one routine edit, one bug with a repro, one cross-file change and one task where the model should ask for clarification. Give each model the same starting context and a fresh conversation.

1. 1Write the acceptance check before the run: test command, expected behavior and files that must not change.
2. 2Run the same prompt and starting context in a fresh conversation for each model.
3. 3Record elapsed time, tokens or cost, test result, files changed and review minutes.
4. 4Reject any run that passes a test by weakening it, changes unrelated files or hides missing evidence behind a summary.
5. 5Choose a default by accepted changes and total review cost, then keep an exception for tasks where another model clearly wins.

Repeat the set when a model, price or Router policy changes. The point is not to crown a permanent winner. It is to keep the team's default tied to current evidence from the work it actually does.