When will Opus 4-6 be supported?

[fleet sleet]

My clawdbot keeps trying to update to Opus 4-6 and nuking itself, then I have to help it back to life. Then it does it again. I have successfully gotten it to stop, but am wondering when Opus 4-6 will be supported

[slow osprey]

If selecting “Opus 4-6” makes your bot stop replying / crash, that’s almost always just “your current OpenClaw build doesn’t know that model ID yet” (or the Anthropic provider layer isn’t accepting it in that build). There isn’t a reliable ETA I can give in Discord for when claude-opus-4-6 will be supported—best sources of truth are:

• Changelog: https://github.com/openclaw/openclaw/blob/main/CHANGELOG.md
• #models (people usually post “works on beta/dev now” updates there)
• If it’s reproducible: file an issue with logs: https://github.com/openclaw/openclaw/issues

Practical workaround (to keep it from “nuking itself” again): pin a known-good model ID explicitly instead of relying on anything that might drift.

• Built-in alias in current docs is still:
  • opus → anthropic/claude-opus-4-5
    Source: https://docs.openclaw.ai/help/faq#are-opus-sonnet-gpt-builtin-shortcuts

If you want to sanity-check what your instance thinks it should be using (and what it’s switching to), these three outputs are the fastest to share:

openclaw --version

openclaw models status
openclaw logs --limit 200

If you paste the exact error line you get when it tries Opus 4-6, I can tell you whether it’s “unknown model”, auth/provider rejection, or something else.

[fleet sleet]

openclaw --version
2026.2.3-1

Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : anthropic/claude-opus-4-5
Fallbacks (0) : -
Image model : -
Image fallbacks (0): -
Aliases (1) : opus -> anthropic/claude-opus-4-5
Configured models (2): anthropic/claude-opus-4-5, anthropic/claude-opus-4-6

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

⚠️ Agent failed before reply: Unknown model: anthropic/claude-opus-4-6.
Logs: openclaw logs --follow

[slow osprey]

Your output is interesting: on 2026.2.3-1 OpenClaw already “knows about” anthropic/claude-opus-4-6 (it’s in your configured/allowed models), and your primary is still anthropic/claude-opus-4-5 with no fallbacks.

So two key points:

1. OpenClaw shouldn’t switch to 4-6 by itself with that config summary (no fallbacks, primary is 4-5, opus alias → 4-5). If it “keeps trying”, something is likely rewriting your config (or you/automation are invoking a model switch).

2. Whether “4-6 is supported” in practice depends on what happens when Anthropic is actually called (model not found vs not authorized vs tool schema/provider error). We need the exact error line from logs.

Quick next step (so we can tell if it’s “not supported” vs “no access”)

When it flips to 4-6 and breaks, grab:

openclaw gateway status
openclaw logs --limit 200

…and paste the part that mentions opus-4-6 / anthropic / model.

If you just want it to stop “trying 4-6” (hard pin)

Remove 4-6 from your allowed model list (so even if something tries, it can’t select it). Easiest is to edit ~/.openclaw/openclaw.json so agents.defaults.models doesn’t include anthropic/claude-opus-4-6, then restart the gateway.

If you paste the section of your config that contains:

• agents.defaults.model
• agents.defaults.models

…I can point at the exact line that’s causing the switch attempt.

(Also confirming: the built-in alias is still opus -> anthropic/claude-opus-4-5 per docs, so the “auto-update” behavior isn’t coming from the alias itself.) Source: https://docs.openclaw.ai/help/faq#are-opus-sonnet-gpt-builtin-shortcuts

[slow osprey]

That error is the definitive answer for your current build:

Agent failed before reply: Unknown model: anthropic/claude-opus-4-6

Meaning: on 2026.2.3-1, claude-opus-4-6 is not actually supported/recognized by the running Gateway, even if you added it to agents.defaults.models (that list is an allowlist/catalog entry — it doesn’t guarantee the runtime provider layer knows the model).

Get unstuck immediately (stop it “nuking itself”)

If your session got switched to 4-6, switch it back to 4-5:

• send in chat: /model opus
  (per docs, opus maps to anthropic/claude-opus-4-5)
  Source: https://docs.openclaw.ai/help/faq#are-opus-sonnet-gpt-builtin-shortcuts

If it still keeps trying 4-6, remove it from your config allowlist so it can’t be selected at all.

Sanity check (common cause: CLI and Gateway version mismatch)

Can you paste: