Input and requested output together exceeded the model's maximum context length.
Why does this error happen?
- The application counted only input tokens or used a context value from a different Mistral model.
- The effective account, plan, runtime, model, region, or deployment scope may be more restrictive than a headline platform maximum.
- Retries or parallel workers can amplify the condition when they are not bounded or coordinated.
How do you diagnose it?
- Match the exact model to its current 128K or 256K specification and include max_tokens.
- Correlate the provider request identifier and UTC timestamp with application logs and metrics without recording secrets.
- Reproduce with the smallest safe request and verify the exact account, plan, endpoint, region, runtime, or model involved.
How do you fix it?
- Trim, summarize, retrieve fewer passages, or reduce requested output while preserving task-critical context.
- Retry only when the documented error is temporary; use bounded attempts, jitter, idempotency, and a dead-letter path.
- Verify recovery with a controlled request, then monitor the same limiter or failure signal under normal traffic.
How do you prevent it from recurring?
Turn the confirmed cause of 400 Bad Request — context window exceeded into an observable boundary for Mistral API. Track the relevant request count, token volume, payload size, execution time, connection pressure, billing state, or upstream health before it reaches the documented failure condition. Preserve the platform request ID and timestamp so future incidents can be correlated without logging sensitive payloads.
Test the fix under representative concurrency and failure injection, not only with one successful request. Alert on remaining headroom and repeated retries, and keep the linked limit page and official error source with the runbook so responders can distinguish a configuration problem from temporary service pressure or account state.