Budgets and limits
Plans
Every plan is a fixed monthly price with a fixed monthly compute budget. Each request counts against the budget at the model's per-token rate. There are no 5-hour or weekly windows inside the month.
| Plan | Price per month | Monthly compute budget |
|---|---|---|
| Starter | $19 | $19 |
| Economy | $39 | $39 |
| Standard | $89 | $89 |
| Pro | $249 | $249 |
| Pro Plus | $499 | $499 |
| Growth | $999 | $999 |
| Scale | $2,499 | $2,499 |
Companies use Teams, which is free, and each API key can carry any plan above with no fee on plans. Enterprise ($1,499/month) adds every Enterprise feature. See pricing for current plans. In a company workspace, each API key carries its own plan, approved by an admin.
Monthly budget
Track your budget live on the Usage page.
Requests run at full speed until the budget is used. They are not slowed down as you use more of it. When the budget runs out, requests return 402 with code budget_exhausted until the period renews. The x-sc-budget-renews-at response header tells you when that is.
There are no overage charges. Nothing is billed on top of your plan price unless you turn on Pay as you go yourself.
To keep working before the renewal:
- Upgrade your plan. An upgrade takes effect immediately.
- On Economy and above, turn on Pay as you go in the dashboard. Compute past your plan is then added in prepaid steps at the usage rate plus 5%, starting at 90% of your monthly budget.
Smart routing uses efficient models for simple work, so your budget lasts longer than if every request is pinned to a premium model.
Smart pacing (retired)
Smart pacing, which slowed responses down instead of stopping at the budget, was retired on October 8, 2026. It is not available on new keys or any current plan. A small number of existing keys that were already using it keep it for now. See Smart pacing was retired.
Rate limits
A high anti-abuse request-rate ceiling applies to every key. If an agent bursts past it, you get 429 with code rate_limit_exceeded and a retry-after header. Wait that many seconds before retrying. Standard OpenAI-compatible SDKs retry this automatically.
Request limits
| Limit | Paid plans | Free trial |
|---|---|---|
| Input per request | about 850,000 tokens | about 500,000 tokens |
| Output per request | 50,000 tokens | 32,768 tokens |
| Images per request | 32, up to 10 MB each | 32, up to 10 MB each |
If you ask for more output than the limit, it is reduced to the limit. Oversized input is trimmed where possible; if it still doesn't fit, the request returns 400.
Fair use
Plans are intended for individual developers and their agents. Occasional bursts are expected. See the fair use policy for details.