OpenClaw Model Down? /model Command and Failover Complete Guide

This post is currently available in Chinese only. Read in Chinese →

When OpenClaw returns request ended without sending any chunks, the model channel is unstable — not OpenClaw itself. Here's how to recover quickly and set up automatic failover.

Three Approaches

Hard switch — SSH to server, edit ~/.openclaw/openclaw.json, change the apiKey or default model ID, restart the gateway. Error-prone and slow.

/model command — Switch models live in chat without touching config or restarting:

/model aigocode-gpt/gpt-5.3-codex   # switch to Codex
/model cc/claude-opus-4-6            # switch back to Claude
/model reset                         # back to config default

Only works with models in your agents.defaults.models whitelist. Set model aliases for one-word switching: /model opus, /model codex.

Fallback chains — Automatic switching with no manual intervention:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "cc/claude-opus-4-6",
        "fallbacks": ["aigocode-gpt/gpt-5.3-codex", "aigocode-gpt/gpt-5.2-codex"]
      }
    }
  }
}

When the primary model times out or hits rate limits, OpenClaw automatically tries the next fallback. New sessions revert to the primary once it recovers.

The Setup

Register multiple providers with different channels in models.providers. Use /model for immediate switching, fallback chains for automatic recovery. Running two providers from different channels means they almost never fail simultaneously.