Which AI Models Does OpenClaw Support? Full 2026 List

OpenClaw Supported Models Overview

OpenClaw uses a model-agnostic design, meaning you can configure any supported AI model and switch freely between them. The framework supports three types of model access:

1. Cloud API models: Claude, GPT-4, Gemini, DeepSeek, MiniMax, etc. — charged by the model provider based on usage
2. Local models (Ollama): Fully offline, no API fees, runs on your own hardware
3. Multi-model failover: Configure multiple models as primary + backup for automatic switching

Full Supported Model List (2026)

Anthropic — Claude Series

Best for: Complex reasoning, code generation, long document analysis, multilingual tasks

Access: Direct API, no China VPN required if using a compliant domestic relay service

OpenAI — GPT Series

Best for: General-purpose tasks, multimodal inputs (images + text), instruction following

Access: Requires API Key, direct access in China is not available without a relay

Google — Gemini Series

Best for: Multimodal tasks, long document analysis, Google Workspace integration

DeepSeek — Domestic Accessible

Best for: Users in China (direct access, no VPN), code generation, Chinese content, cost-sensitive workloads

Access: Direct connection in China, generous free tier

MiniMax — Domestic Accessible

Best for: China users getting started (free quota upon registration), Lesson 01 default recommendation

Access: Direct connection in China, free quota on registration — ideal for getting started

Ollama — Local Models (Fully Offline)

Ollama lets you run open-source models locally on your Mac, Linux, or Windows machine (via WSL2):

Best for: Users with strict privacy requirements, scenarios where data must not leave the device, zero API cost

How to Configure Models

Add model providers in openclaw.json:

{
  "models": {
    "mode": "merge",
    "providers": {
      "anthropic": {
        "apiKey": "${ANTHROPIC_API_KEY}",
        "models": [
          { "id": "claude-sonnet-4-6", "name": "Claude Sonnet 4.6" }
        ]
      },
      "minimax": {
        "baseUrl": "https://api.minimax.io/anthropic",
        "apiKey": "${MINIMAX_API_KEY}",
        "api": "anthropic-messages",
        "models": [
          { "id": "MiniMax-M2.1", "name": "MiniMax M2.1" }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "minimax/MiniMax-M2.1",
        "fallback": ["anthropic/claude-haiku-4-5"]
      }
    }
  }
}

See Lesson 05: Multi-Model & Failover for detailed configuration.

How to Choose the Right Model

FAQ

How often is the model list updated?

The OpenClaw team typically adds support for major new models within 1–2 weeks of release. The community also contributes adapters for smaller models.

How does model billing work?

OpenClaw itself doesn't charge for model usage. You pay the model provider directly (Anthropic, OpenAI, etc.) based on your actual API usage. Ollama local models are completely free.

Can I limit which model a specific skill uses?

Yes. You can specify a model at the skill level in OpenClaw's configuration, so different skills use different models optimally (e.g., light tasks use a cheap fast model, complex reasoning uses a more capable model).