GitHub Issue: Model Fallback Not Triggering on Rate Limit (429)

Repository: https://github.com/punkpeye/openclaw/issues/new

Bug Report

Description

Model fallback chain does not trigger when the primary model returns a 429 (rate limit) error. The error message shows "All models failed (1)" even when multiple fallback models are configured.

Expected Behavior

When the primary model (e.g., google/gemini-3-flash-preview) returns a 429 rate limit error, OpenClaw should automatically try the next model in the fallbacks array (e.g., anthropic/claude-haiku-4-5).

Actual Behavior

OpenClaw attempts only the primary model, fails with 429, and returns:

Agent failed before reply: All models failed (1): google/gemini-3-flash-preview: Provider google is in cooldown (all profiles unavailable) (rate_limit).

The "(1)" indicates only 1 model was attempted, not the 4 configured.

Configuration

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "google/gemini-3-flash-preview",
        "fallbacks": [
          "anthropic/claude-haiku-4-5",
          "moonshot/kimi-k2.5",
          "deepseek/deepseek-chat"  
        ]
      }
    }
  },
  "auth": {
    "profiles": {
      "google:default": { "provider": "google", "mode": "api_key" },
      "anthropic:default": { "provider": "anthropic", "mode": "api_key" },
      "moonshot:default": { "provider": "moonshot", "mode": "api_key" },
      "deepseek:default": { "provider": "deepseek", "mode": "api_key" }
    }
  }
}

All 4 providers have valid API keys configured in auth-profiles.json.

Logs

No fallback attempt is visible in logs. Only the primary model failure:

[diagnostic] lane task error: lane=main durationMs=15388 error="FailoverError: LLM error: {
  "error": {
    "code": 429,
    "message": "Resource has been exhausted (e.g. check quota).",
    "status": "RESOURCE_EXHAUSTED"
  }
}"

Environment

• OpenClaw version: 2026.2.6-3
• Platform: Railway (Docker)
• Node.js: 22.22.0

Workaround

Currently using a provider with higher rate limits (e.g., Anthropic or DeepSeek) as primary model.

Steps to Reproduce

1. Configure primary model with a provider that has strict rate limits (e.g., Google Gemini free tier)
2. Configure fallback models with different providers
3. Ensure all providers have valid API keys
4. Send messages until primary model hits rate limit
5. Observe that fallback models are not attempted