openclaw - 💡(How to fix) Fix v2026.3.28: Rate-limited primary model causes LiveSessionModelSwitchError cascade, fallbacks never used [1 participants]

[model-fallback/decision] candidate_failed requested=google/gemini-3-flash-preview candidate=anthropic/claude-haiku-4-5 reason=unknown [agent/embedded] live session model switch detected: anthropic/claude-haiku-4-5 -> google/gemini-3-flash-preview [openai-compat] chat completion failed: LiveSessionModelSwitchError: Live session model switch requested

TL;DR

The most likely fix is to modify the fallback chain engagement logic to properly handle rate-limited primary models.

Guidance

• Verify that the fallback candidates are being correctly configured and prioritized in the agent settings.
• Investigate the LiveSessionModelSwitchError and candidate_failed reason=unknown errors to determine why the fallback candidates are being rejected before the API call.
• Consider implementing a retry mechanism with exponential backoff for the primary model to prevent overwhelming the API with repeated requests.
• Review the session lifecycle and model switching logic to ensure it is correctly handling rate-limited primary models and engaging the fallback chain.

Example

No code snippet is provided as the issue does not contain sufficient information to create a specific example.

Notes

The issue seems to be specific to the HTTP API endpoint and does not affect the Telegram integration, suggesting that the problem may be related to the session lifecycle or model switching logic in the HTTP endpoint.

Recommendation

Apply a workaround by switching the primary model to a paid model to prevent rate limiting, as described in the issue's workaround section, until a more permanent fix can be implemented.

FAIL-SAFE

If the issue persists, try to isolate the problem by testing the fallback chain with a different primary model or by reducing the number of fallback candidates to see if the issue is related to a specific model or configuration.