Quick answer
Grok documents per-model rate dimensions as RPS and TPM vary by model and cumulative-spend tier. The per-second request limit is derived from the minute budget and prevents full-minute bursts.
Verified Aug 22, 2026Official source
Current limits
| Constraint | Current value | Scope | Verified source |
|---|---|---|---|
| per-model rate dimensionsThe per-second request limit is derived from the minute budget and prevents full-minute bursts. | RPS and TPM vary by model and cumulative-spend tier | Per xAI API team | xAIAug 22, 2026 |
Why does this limit matter?
A workload must fit both token throughput and request pacing for the selected model.
This value is scoped to Per xAI API team; a different plan, runtime, model, endpoint, region, or account can produce a different effective constraint.
What should you check?
- Open the xAI Console Rate Limits page for the team and model.
- Confirm the exact plan, model, runtime, endpoint, region, and account that serve the failing workload.
- Record the observed value, response headers or configuration, timestamp, and source without logging secrets.
Important caveats
- Voice and Imagine APIs use separate limits and increase paths.
- Treat the official source and live account configuration as authoritative if they differ from this verified snapshot.
HyperObserve reports the documented platform constraint. Your application, SDK, gateway, provider, region, or account can impose a lower effective limit.
Related