Limits
Limits protect shared capacity. Current values for your organization are shown in the console.
Request limits
| Limit | Where it’s set | When exceeded |
|---|---|---|
| Requests per minute | Per deployment, by you | 429 rate_limited |
| Concurrent requests | Per organization, by plan | 429 rate_limited |
| Maximum output tokens | Per deployment, by you | Clamped silently to the maximum |
| Context length | Per base model | 400 context_length_exceeded |
| Request body size | 1 MB | 400 |
| Sandbox requests | Per organization per day | 429 rate_limited |
Plan limits
Each plan limits adapters, active deployments, clients and team members (including pending invitations); see Pricing. The sandbox uses the Developer plan’s limits. Archived items don’t count. Going over a limit returns 403 plan_limit_reached from the console API.
Known limitations in V1
- Only chat completions with text content. No tool/function calling, images,
n> 1,logprobsor structured output modes. - No
/v1/completionsor/v1/embeddingsendpoints. - LoRA adapters only; no full fine-tuned checkpoints, DoRA or adapters that change the vocabulary.
- All plans run on shared capacity. The first request to an adapter that hasn’t been used recently can take longer while it loads.
- No dedicated capacity, regions or uptime commitment during the preview.