Repository navigation
[Bug]: Model rate limit error causes immediate termination instead of auto-retry #16994
Description
Activity
This issue might be a duplicate of existing issues. Please check:
- [Feature]: Auto-retry on MCP tool rate limit instead of terminating #16993: Same error message ("You have exceeded the request rate limit") and same request for auto-retry with backoff — filed by the same author, now closed. This issue may be a continuation or broader version of that one.
- LLM provider SDK lacks retry logic for transient API errors #11705: Covers the same missing retry/backoff logic for LLM provider SDK transient errors including rate limits.
- More resilient retry policy for Amazon Bedrock rate limiting #452: The original retry policy issue for rate limiting that is already referenced in this report.
Description
When the model provider returns the following rate limit error, opencode terminates the current task immediately instead of retrying automatically:
Expected Behavior
Detect this rate limit response and apply retry-with-backoff logic, similar to what was added for other rate limit cases (see #452). The user should not need to restart their entire task.
Current Behavior
opencode aborts the task immediately upon receiving this error message. There is no retry attempt.
Steps to Reproduce
Trigger a request that causes the model provider to respond with the rate limit message above (e.g. by sending many requests in a short period).
Notes
You have exceeded the request rate limitmay not be matched by the current retry detection logic, causing it to fall through as a fatal error