(MOT-3929) fix(provider-anthropic): surface adaptive thinking by default on thinking-capable models

[andersonleal]

Summary

• With no explicit thinking_level, provider-anthropic always sent config: None — thinking-capable models (Sonnet 4.6+, Opus 4.7+, Fable/Mythos 5) never got adaptive thinking unless the caller explicitly requested a level. provider-xai surfaces reasoning by default with no explicit level; this brings provider-anthropic to parity.
• build_thinking_config now requests ADAPTIVE (server default effort, no output_config) when no level is given and the model's supports_thinking is explicitly Some(true). Unknown (None) or explicitly Some(false) models keep the old behavior, so the implicit default can never 400 a non-thinking or unknown model.
• Cargo.lock churn across approval-gate/other providers is from the harness v1.1.12 bump (iii-sdk/iii-helpers version resolution), not a functional change.

Test plan

• cargo test thinking in provider-anthropic (14 passed)
• Updated/added unit tests: absent_level_defaults_on_for_thinking_models, absent_level_stays_off_without_thinking_support

https://linear.app/motia/issue/MOT-3929/provider-anthropic-surface-adaptive-thinking-by-default-on-thinking

Summary by CodeRabbit

• Bug Fixes

  • Updated thinking settings so supported models now automatically use adaptive thinking when no level is specified.
  • Kept thinking turned off for models that do not support it or when support is unknown.
  • Ensured the default behavior preserves empty warnings and no extra effort settings when thinking is not enabled.

• Tests

  • Added coverage for both supported and unsupported model cases to verify the new default behavior.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
workers	Ready	Preview, Comment	Jul 9, 2026 4:30pm
workers-tech-spec	Ready	Preview, Comment	Jul 9, 2026 4:30pm

[github-actions]

skill-check — worker

0 verified, 41 skipped (no docs/).

Layer	Result
structure	✓
vale	✓
ai	✓
render	✓

Four for four. Nicely done.

[coderabbitai]

Warning

Review limit reached

@andersonleal, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 49 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: b045dbf9-6e55-4ef2-aa9d-affe4cf9fa02

📥 Commits

Reviewing files that changed from the base of the PR and between 751c30e and 724e9a8.

📒 Files selected for processing (2)

• .github/workflows/ci.yml
• provider-openai-codex/src/stream_fn.rs

📝 Walkthrough

Walkthrough

Modified build_thinking_config in provider-anthropic to default to adaptive thinking when thinking_level is absent and the model explicitly supports thinking (supports_thinking == Some(true)); otherwise config remains disabled. Updated tests to cover both branches, replacing the previous single "absent means off" test.

Changes

Adaptive Thinking Default Behavior

Layer / File(s)	Summary
Conditional adaptive default and tests provider-anthropic/src/thinking.rs	Absent thinking_level now yields config: Some(ADAPTIVE) when the model supports thinking, else config: None; effort stays None and warnings empty. Tests split into two cases covering supported and unsupported/unknown models.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Poem

A rabbit peeks at models near,
"Do you think?" it asks with cheer.
If yes, adaptive thoughts unfold,
If not, the config stays cold.
Two tests now hop where one once stood— 🐇✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: enabling adaptive thinking by default for thinking-capable Anthropic models.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch andersonleal/mot-3929-provider-anthropic-surface-adaptive-thinking-by-default-on

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}