Teams
Cursor Token Rate: The Fee Explained (and How to Avoid It)
The Cursor Token Rate is a 25-cents-per-million-tokens charge that Teams and Enterprise accounts pay only on non-Auto requests to third-party models. Auto requests and first-party models (Composer 2.5, Grok 4.5) are exempt. It counts input, output and cached tokens, and applies across included usage, on-demand usage and BYOK.
On this page
What is the Token Rate on my Cursor bill?
It is a flat infrastructure fee: $0.25 per million tokens, charged only on non-Auto, third-party model requests for Teams and Enterprise customers. Cursor describes it as covering "search infrastructure, custom model execution and routing, processing and infrastructure costs" — the machinery around the model call, not the model tokens themselves, which bill at their own rates.
- Request
- Auto (Cursor picks the model)
- Token Rate applies?
- No — exempt regardless of which model Auto routes to
- Request
- 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. selected explicitly
- Token Rate applies?
- No — first-party models are exempt
- Request
- Grok 4.5 selected explicitly
- Token Rate applies?
- No — first-party models are exempt
- Request
- Claude / GPT / Gemini selected explicitly
- Token Rate applies?
- Yes — third-party model outside Auto
- Request
- Third-party model via BYOK
- Token Rate applies?
- Yes — BYOK does not exempt the fee
| Request | Token Rate applies? |
|---|---|
| Auto (Cursor picks the model) | No — exempt regardless of which model Auto routes to |
| 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. selected explicitly | No — first-party models are exempt |
| Grok 4.5 selected explicitly | No — first-party models are exempt |
| Claude / GPT / Gemini selected explicitly | Yes — third-party model outside Auto |
| Third-party model via BYOK | Yes — BYOK does not exempt the fee |
Per cursor.com/help/models-and-usage/token-rate, checked 2026-07-16.
Exempting Auto and first-party models makes the fee a routing incentive: teams that pin every request to a named third-party model pay for the privilege, teams that let Auto route (or standardize on ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. / Grok 4.5) don't. Knowing that is half of managing it.
This is covered hands-on in Teams and Enterprise Admin — 6 short modules, free to read.
Which tokens does it count?
All of them: for each eligible request the charge covers input tokens, output tokens and cached tokens, and it applies across all three usage categories — included usage, on-demand usage and BYOK. Two consequences follow that surprise teams:
- Cache hits are not free for this fee — cached tokens are cheap on the model side but still count toward the Token Rate, so long conversations with big cached prefixes accumulate it steadily.
- Large-context work multiplies it. A big context window means more input tokens per turn; an agent session is many turns. The same mechanics that make Max Mode-style large-window requests expensive make this fee visible on heavy third-party usage.
- One 200k-token agent turn
- ≈ half a cent of Token Rate (0.2M × $0.25/M).
- A 40-turn agent session at that size
- ≈ 20 cents — on top of model token costs.
- A team of 25 doing that daily
- ≈ $100/month of Token Rate alone — the line item people write in asking about.
Illustrative arithmetic on the published rate, not a quote; verify current terms on Cursor's pricing docs.
How do I reduce or avoid the Token Rate?
- 1Default to Auto. Cursor's stated avoidance path is exactly this: "use Auto or a first-party model." Auto-routed requests are exempt even when Auto lands on a third-party model.
- 2Standardize everyday work on first-party models. 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 speed, Grok 4.5 for hard tasks — both exempt. Reserve named third-party models for the cases where a specific model demonstrably wins.
- 3Audit who is pinning models. Usage dashboards show which teammates run non-Auto third-party requests; often it's a leftover personal default, not a considered choice.
- 4Don't assume BYOK dodges it. Bringing your own key moves the model bill to your provider account but the Token Rate still applies to those requests.
Billing mechanics are the fastest-moving part of Cursor's product surface — this year alone repackaged Max Mode twice. Confirm the current rate and exemption list at cursor.com/help/models-and-usage/token-rate before writing it into a budget or policy.
Frequently asked questions
What is the Token Rate charge on my Cursor invoice?
A flat $0.25 per million tokens that Teams and Enterprise accounts pay on non-Auto requests to third-party models. It covers Cursor's search, routing and infrastructure costs and counts input, output and cached tokens.
Does the Token Rate apply to Auto requests?
No. Auto requests are exempt, as are first-party models like Composer 2.5 and Grok 4.5. The fee applies only when someone explicitly selects a third-party model.
Do I pay the Token Rate with my own API keys (BYOK)?
Yes. BYOK moves model token costs to your provider account, but the Token Rate still applies to non-Auto third-party requests made through Cursor.
Do individual Pro plans pay the Token Rate?
Cursor documents it as a Teams and Enterprise charge structure. Individual plans meter model usage differently; check your plan's usage docs rather than assuming either way.
Are cached tokens really charged?
For this fee, yes — the charge explicitly encompasses input, output and cached tokens. Caching still saves you on the model-side rates; it just doesn't reduce the Token Rate.
Sources & last verified
Cursor ships frequently. Facts verified against primary sources on July 16, 2026.