Choose a route before chasing the latest model
Before asking "Claude or GPT?" or "Kimi or GLM?", answer these in order:
- Which provider route can you maintain?
- What complexity level do you need?
- Which specific model fits that route?
Model configuration commands
openclaw config set model.provider deepseek
openclaw config set model.api_key sk-your-key-here
openclaw config set model.name deepseek-chat
Or for OpenAI:
openclaw config set model.provider openai
openclaw config set model.api_key sk-***
openclaw config set model.name gpt-4o
Supported providers
| Provider | Typical use |
|---|---|
| DeepSeek | Cost-effective, strong for coding and Chinese |
| OpenAI | GPT family, broad tooling |
| Claude | Anthropic models, long context |
| Zhipu | GLM models, Chinese workflows |
| Kimi | Moonshot models, long context |
| Tongyi | Alibaba models, Chinese enterprise |
Check the official docs for the latest model names and API fields.
Fallback concept
When the primary model fails or hits limits, OpenClaw can fall back to another provider. Configure fallback only after a single model works reliably. See Model router for details.
Verify your setup
After configuring the model:
openclaw doctor
Ensure the model check passes. If it fails, verify the API key, network, and provider name.
Recommended order
- Pick one provider you can keep using
- Get a single model working end-to-end
- Add fallback only if you need resilience
- Add routing only when you need task-based switching
Where to go next
- For fallback and routing: Model router
- For cost control: Model cost control
- Official docs → docs.openclaw.ai
- Issues → GitHub Issues