Fix: Gemini Models Return 400 (Request Formatting Issue)

Google Gemini uses a different API request format from OpenAI. If OpenClaw sends an OpenAI-formatted request to the Gemini endpoint — because the model name lacks the google/ prefix or the provider is misconfigured — Gemini returns a 400 error. This guide covers all the config fixes to get Gemini working correctly.

Step 1: Fix the Model Name Prefix

The google/ prefix tells OpenClaw to use the Gemini request formatter. Without it, OpenClaw may assume OpenAI format and produce a 400 error:

{
  "model": "google/gemini-2.0-flash"

  // NOT: "gemini-2.0-flash" (missing prefix — uses wrong formatter)
  // NOT: "gemini/gemini-2.0-flash" (wrong prefix)
}

Available Gemini model names with correct prefixes:

Step 2: Explicitly Configure the Google Provider

{
  "model": "google/gemini-2.0-flash",
  "providers": {
    "google": {
      "enabled": true,
      "baseUrl": "https://generativelanguage.googleapis.com/v1beta"
    }
  }
}

Step 3: Add a Google AI Studio API Key

The CLI OAuth flow can cause authentication mismatches that lead to 400 errors. Using a direct API key is more reliable:

{
  "model": "google/gemini-2.0-flash",
  "credentials": {
    "google": {
      "apiKey": "AIza..."
    }
  }
}

Step 4: Verify Gemini is Working

# Test your API key directly
curl -s "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key=YOUR_KEY"   -H "Content-Type: application/json"   -d '{"contents":[{"parts":[{"text":"Hello"}]}]}'

# If this succeeds, restart OpenClaw gateway
openclaw gateway restart