Cursor model pricing is the usage-based layer of Anysphere’s licensing. Each paid plan includes a monthly allowance of model usage, and what happens inside and beyond that allowance depends on which model a request uses, which plan the account is on and whether the customer supplied its own provider key. The Pricing Policy defines the fee mechanics: Model API Fees are Total Tokens multiplied by the applicable Model Pricing rate, and Cursor Token Fees are Total Tokens multiplied by the Cursor Token Rate.[3]
Editions and pools
Docs describe two separate usage pools, each resetting with the monthly billing cycle. The Cursor Models pool covers Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5 and offers significantly more included usage for those models. The Other Models pool is for third-party models, charged at the model’s API price. Pro, Pro Plus and Ultra include both pools, with the option to pay for additional usage; the Start plan includes only the Cursor Models pool.[1] The pools are visible in the editor settings and on the usage dashboard.[1] Teams and Enterprise seats are also allocated both pools.[2]
The catalog rows are Cursor Models pool and Other Models pool.
Metrics: price per million tokens
All prices in the Docs are per million tokens, split into input, cache write, cache read and output. The selection below reproduces the rates published at retrieval.[1]
| Model | Pool | Input | Cache write | Cache read | Output |
|---|---|---|---|---|---|
| Grok 4.7 | Cursor Models | $2 | none listed | $0.5 | $6 |
| Grok 4.7 (Fast) | Cursor Models | $4 | none listed | $1 | $12 |
| Composer 2.5 | Cursor Models | $0.5 | none listed | $0.2 | $2.5 |
| Composer 2.5 (Fast) | Cursor Models | $3 | none listed | $0.5 | $15 |
| Claude Fable 5.1 | Other Models | $10 | $12.5 | $0.25 | $50 |
| Claude Opus 5.5 | Other Models | $4 | $5 | $0.2 | $20 |
| Claude Sonnet 5.5 | Other Models | $2 | $2.5 | $0.1 | $10 |
| Claude Haiku 5.5 | Other Models | $0.1 | $0.125 | $0.01 | $0.5 |
| Gemini 3.1 Pro | Other Models | $2 | none listed | $0.2 | $12 |
| GPT-5.6 Sol | Other Models | $4 | $5 | $0.4 | $20 |
| GPT-5.6 Terra | Other Models | $2 | $2.5 | $0.2 | $12 |
The Docs list more models than shown here, including Fast and 500k-context variants of Grok 4.7, which are priced higher than the base model.[1] Model Pricing is subject to change in accordance with the model provider’s list pricing.[3] Because different models have different API costs, model selection changes how quickly the included usage is consumed.[1] Total Tokens is defined as the aggregate tokens processed across input, output and cache read and write tasks.[3]
Counting: the Cursor Token Rate
On Teams and Enterprise plans, third-party model requests include a Cursor Token Rate of $0.25 per million tokens. It applies on top of model API pricing for included usage, on-demand usage and bring-your-own-key usage, and it applies when Auto routes to a third-party model. First-party Cursor models, including Grok and Composer, are exempt.[1] The Docs describe it as covering custom model execution (Tab, Apply and similar features) and infrastructure and processing costs, and as applying to input, output and cached tokens.[2] The Pricing Policy lists Cursor Token Fees alongside Model API Fees as the only elements eligible to count toward Enterprise Precommitted Usage.[3]
A worked illustration using only published rates (commentary, not a quotation): a Teams request to Claude Sonnet 5.5 with 1,000,000 input tokens and 200,000 output tokens has model cost of $2 plus $2, or $4. The Cursor Token Rate applies to the 1,200,000 tokens at $0.25 per million, adding $0.30. The same request to Composer 2.5 would carry no Cursor Token Rate.
Auto and Cursor Router
Auto is Cursor’s automatic model selection, run by Cursor Router, a classifier that sends each request to a model suited to the task.[4] It has three modes, Cost (the former Auto), Balance (the default for new users) and Intelligence.[4] All Auto modes bill at the list price of the model each request is routed to, so a request routed to a third-party model draws from the Other Models pool and, on Teams and Enterprise, incurs the Cursor Token Rate.[1][6] Subagents can also run a named third-party model while the picker still shows Auto, Grok or Composer, and those requests bill the Other Models pool.[6]
Router launched for Teams and Enterprise first; individual plans were to receive it a few months later, and Enterprise teams start with it off until an admin opts in.[11] Admins can enable or disable it for the team or specific groups, choose which optimization modes members see, and enforce Auto softly (new chats default to it) or hard (the model picker is locked).[10] Anysphere’s Overages page, answering why costs increased after the Router launch, states that all Auto modes bill at the routed model’s list price and that users of the former Auto mode were defaulted to Cost mode.[9]
Bring your own key
Customers can add their own API keys for OpenAI, Anthropic, Google, Azure or AWS Bedrock models. On individual plans, requests made with the customer’s own key do not draw from included usage, because the provider bills the model cost.[5] On Teams and Enterprise, every third-party request still carries the Cursor Token Rate, which is drawn from the Other Models allowance and billed as on-demand usage when that allowance is exhausted; if on-demand usage is off, such requests stop until the next billing cycle.[5] Custom keys work only with chat models, and Tab completion continues using Cursor’s built-in models.[5] The key is sent to Cursor’s backend with every request, and Cursor’s zero data retention policy does not apply to such requests.[5]
On-demand usage and overage
Beyond the included pools, usage continues at the same API rates with pay-as-you-go billing, billed monthly, and requests are not downgraded in quality or speed.[1] Overage is billed at API rates with no markup, plus the Cursor Token Rate where it applies.[9] When a Teams member exhausts their Other Models usage, Cursor first moves them to the Cursor Models pool, and on-demand usage begins only when both are consumed or when they keep using third-party models with on-demand enabled.[2]
Cloud and data residency
Opting in to US-only data residency incurs a 10% uplift on Model pricing for eligible models, and it is available to Enterprise customers enabled per team.[7] Bring-your-own-key use is outside the data residency guarantee.[7] Max Mode, which extends a model’s context window, is available only on legacy request-based plans and is billed at the model’s API rate plus 20%.[8]
Out of scope
This article does not list every model price, does not reproduce pool sizes (which are not published as fixed figures in the documents retrieved), and does not cover the terms of the underlying model providers, whose own pricing and data policies apply to their models. Subscription pricing, Bugbot and Cloud Agent billing are covered in the related articles.