Autumn 2026 Models in OpenClaw

Release notes named several chat models between v2026.9.2 and v2026.9.9. This page is how to attach them. For which one should be default, use Choosing LLMs.

Model IDs in config are version-specific. If a name 404s, list models from your install (Control UI or the official CLI for your build) instead of copying a blog string.

Named in official notes

VersionModels called out
v2026.9.2OpenAI GPT-6 Astra; Meta Muse Spark 1.3
v2026.9.6Claude Opus 5.5; GPT-6 Sol and Luna; Grok 4.7; optional Decision Models
v2026.9.7OpenAI Agents API; Sign in with ChatGPT (Beta)—account path, not a model ID
v2026.9.9GPT-6.1 Sol in Codex; Claude Haiku 5.5

Wiring order

  1. Set the provider keys first: Anthropic, OpenAI, plus xAI/Google/DeepSeek as needed.
  2. Pick one default and one fast model. Do not point every specialist at Opus 5.5.
  3. Restart or reload so the Gateway sees the new IDs. Test with a one-line prompt, then a tool call.
  4. Watch 30-day Usage for a day before you roll it to all channels.

Sign in with ChatGPT

v2026.9.7 lists Sign in with ChatGPT as a beta. That is an identity/billing path, not a replacement for a self-hosted API key. Keep Gateway bind on localhost either way. Dedicated page: Sign in with ChatGPT. Hosted Codex harness: Agents API. Typed judgments: Decision Models.

Related