Rate limits and quotas
Hourly rate limits that protect the service, monthly plan quotas, and how to see where you stand.
Two kinds of limit apply. Rate limits are sliding one-hour windows per organization that protect the service; exceeding one returns 429 rate_limited with a Retry-After header, and the Python client retries automatically after that delay. Quotas are plan allowances; exceeding one returns 403 quota_exceeded and does not resolve by waiting a minute.
| Rate limit | Per hour on Free | Counts |
|---|---|---|
Models | 40 | POST /v1/models, POST …/versions and forks into the organization |
Predictions | 600 | POST /v1/predictions, …/batch (one hit per batch) and …/predict |
Explanations | 120 | POST /v1/explanations |
Analyses | 30 | POST /v1/analyses |
Tensor uploads | 20,000 | PUT …/structure/tensors/{key}; a large model is thousands of tensors |
API keys | 20 | POST /v1/api-keys |
Invitations | 30 | POST /v1/organization/invitations and …/resend |
Organizations | 5 per account | POST /v1/organizations. An account can own 10. |
Billing links | 30 | POST /v1/billing/checkout and …/portal together; …/refresh 120 |
Sign-in / sign-up | 20 / 5 per IP | Login and signup; verification emails 5 per IP and 3 per address |
The first seven limits count per organization and grow with its plan: three times the figures above on Standard, ten times on Team and thirty times on Enterprise. Organizations, billing links and sign-in do not change with the plan.
Quotas on the free plan: 10 models, 5 API keys, and per calendar month 5,000 logged predictions, 500 explanations and 20 analyses. Counts are taken from stored rows, so deleting a model frees its slot and its logged predictions, and re-running a notebook that replaces its models does not add up. Paid plans raise every allowance and the hourly rate limits; the pricing page has the tiers, and GET /v1/billing lists each plan's limits and changes the plan. Your organization's current limits and this month's usage are in quotas on GET /v1/organization.
An owner or admin can cap the organization below its plan with settings.quotas on PATCH /v1/organization, for example to stop a runaway job; raising a limit above the plan is a plan change.
{
"error": {
"type": "rate_limit_error",
"code": "rate_limited",
"message": "Too many explanations. Try again later.",
"details": {"limit": 120, "window_s": 3600}
}
}{
"error": {
"type": "permission_error",
"code": "quota_exceeded",
"param": "explanations_per_month",
"message": "The free plan allows 500 explanations this month. Upgrade or wait for the next cycle.",
"details": {"used": 500, "limit": 500, "plan": "free"}
}
}{
"plan": "free",
"limits": {
"models": 10,
"api_keys": 5,
"predictions_per_month": 5000,
"explanations_per_month": 500,
"analyses_per_month": 20
},
"usage": {
"models": 4,
"api_keys": 2,
"predictions_per_month": 1560,
"explanations_per_month": 27,
"analyses_per_month": 6
}
}