OpenClaw Model Selection: Which LLM to Use

Choose an OpenClaw model from the job and the data boundary. A fast hosted model may suit a short classification task; a larger model may help with a difficult code review; a local endpoint may fit a workflow that cannot send its input to a hosted provider. The Gateway and the model provider are separate decisions.

Write down the task first

1. Describe the output and the context it needs.
2. Mark whether the task needs tools, files, images, or a long session.
3. Record the latency, quality, and review effort that would make the task useful.
4. Choose a provider and model that meet those constraints, then test with representative but non-sensitive input.

Compare providers without stale claims

Check the provider's current model identifier, context limit, tool support, regional availability, and pricing on its own documentation. Do not copy an old price or assume that a model label has the same capabilities across providers. Keep credentials outside the repository and record which provider handled a test.

openclaw models
openclaw gateway status
openclaw doctor

Test the choice in OpenClaw

Run one read-only task, inspect the answer against its source, and repeat after changing only the model. For a fallback, test authentication, rate-limit, and timeout paths separately because a retry may duplicate a side effect.

Use the official model provider reference for configuration boundaries and the security guide for credentials, tools, and Gateway exposure. Recheck the provider's current terms before making a cost or privacy claim.