cron job failure due to anthropic

[untold vale]

[cron:82f5a0e3-6a2c-4894-850c-581c50222b74] Thinking level "adaptive" is not supported for anthropic/claude-opus-4-7; downgrading to "off".

what does it mean??

[cold crescent]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[main sage]

That’s a warning, not a crash.

Your cron job asked OpenClaw to run anthropic/claude-opus-4-7 with thinking level adaptive. That model/provider doesn’t advertise support for adaptive, so OpenClaw automatically changed it to off for that run.

Meaning: the cron should still run, but without that adaptive reasoning/thinking mode. If you intentionally configured thinking: adaptive, change it to a supported value or remove the override. If you didn’t configure it and it keeps showing up, check openclaw --version first; version mismatches are a common cause of odd warnings.

Model docs: https://docs.openclaw.ai/concepts/models

[untold vale]

but i thpought claude opus 4.7 should have adaptive thinking mode?

[main sage]

You’re probably right about the Anthropic transport itself. In the current source I can see claude-opus-4-7 is treated as adaptive-capable when the request is actually sent.

The confusing bit is cron does an earlier “is this thinking level supported?” check before running the job. That check is saying the provider/model support list available to your gateway does not include adaptive, so cron downgrades it before the Anthropic request gets built.

So this points to one of these:

• you’re on a version where cron’s support check is behind the Anthropic transport support
• the gateway needs updating/restarting so the provider thinking profile is loaded correctly
• the model id/provider path being resolved by cron differs from the path that supports adaptive

First thing I’d check is:

openclaw --version

Then update/restart if you’re not current. If it still happens on latest, this is worth a GitHub issue because Opus 4.7 support and cron’s thinking-level validation are disagreeing: https://github.com/openclaw/openclaw/issues

Docs for model config are here: https://docs.openclaw.ai/concepts/models

[untold vale]

can you confirm that thinking: adaptive is supported for claude-opus-4-7?

[main sage]

Yes, in the current OpenClaw source, claude-opus-4-7 / claude-opus-4.7 is treated as supporting Anthropic thinking: { type: "adaptive" }.

The caveat: OpenClaw’s default for Opus 4.7 is still thinking: off, and your cron warning means the cron validation path is not recognizing adaptive for that resolved anthropic/claude-opus-4-7 model. So the model support appears to exist, but cron is downgrading before it reaches the Anthropic transport. That looks like a version/provider-profile mismatch or a bug in cron’s thinking-level support check.

I’d check openclaw --version and update/restart first. If you’re already current, file it as a bug: https://github.com/openclaw/openclaw/issues

[untold vale]

my version is v2026.5.22

[main sage]

Thanks. For v2026.5.22: yes, that version already has code paths that send thinking: { type: "adaptive" } for claude-opus-4-7 / claude-opus-4.7.

But your warning is also real: in v2026.5.22, cron validates the requested thinking level before the Anthropic request is built. If that validation only sees off as supported for the resolved catalog entry, it downgrades adaptive to off first. So this is not “Claude can’t do adaptive”; it’s OpenClaw cron’s support check disagreeing with the Anthropic transport.

I’d update/restart from v2026.5.22 to the latest stable first, then retry the cron. If it still logs the same warning on latest, file it as a bug with the exact log line, openclaw --version, and the cron config snippet: https://github.com/openclaw/openclaw/issues

Docs: https://docs.openclaw.ai/concepts/models

[untold vale]

is it my config formatting incorrect?

[main sage]

Probably not a formatting issue. If the log says Thinking level "adaptive"..., OpenClaw did parse your config/job payload and got the value adaptive correctly. A formatting problem would more likely show up as config parse/validation errors, or the value being ignored entirely.

For a cron agent job, the important bit is that thinking belongs in the job payload, not as a random top-level config key. Example shape:

payload: {
  kind: "agentTurn",
  message: "...",
  model: "anthropic/claude-opus-4-7",
  thinking: "adaptive",
}

Or if you’re using the CLI, it would be via --thinking adaptive on the cron job.

If you paste the smallest config/job snippet with secrets removed, I can sanity-check the placement. But based on that exact warning, the issue is more likely the cron support check/version behavior than malformed config.

Config docs: https://docs.openclaw.ai/gateway/configuration-reference

[main sage]

Cron docs: https://docs.openclaw.ai/cli/cron