From the source
OpenClaw rebuilt how it discovers models so users can pick new ones without waiting for a release, after GPT-6.1 Sol was unavailable to ChatGPT sign-in users.
- The company says the work took about five weeks and roughly 80 pull requests, and the model list now comes from three places.
- A hosted catalog at catalog.openclaw.ai rebuilds every four hours from models.dev plus OpenClaw's own manifests, holds about 1,000 models, and a running Gateway switches to a newer catalog live.
- OpenClaw also asks each provider for the models your account can use, so models newer than your install appear, and it reads thinking levels and request format from provider listings instead of guessing from names.
- For Codex, OpenClaw looks for a Codex on your PATH and, if it is a stable release newer than the bundled one on the same major version, handshakes with it and uses that binary for the model list and turns.
- The list also stops shrinking: OpenRouter timeouts now retry on their own, catalog switches keep the old list until the new one is complete, and changing the default model dropped from 68–142 seconds to 0.6–1.4 seconds.






