[octo-patch]

Reason: The executable MiniMax catalog still offers legacy IDs, and its response handling drops the thinking blocks required by MiniMax-M3.1-Flash tool conversations.

Changes

• Advertise MiniMax-M3.1-Flash first and MiniMax-M3 second. Keep explicit saved, custom, and legacy model IDs selected and send them unchanged.
• Preserve complete native assistant blocks through tool continuations and subsequent chat turns, including thinking signatures and tool IDs.
• Return text correctly when thinking appears first or streaming has no callback. Leave thinking and effort omitted so the API supplies Flash adaptive/max and M3 thinking-off defaults.
• Keep image/video content blocks intact and update the MiniMax model documentation.

Validation

• ruff check src tests scripts --select=E9,F63,F7,F82 — passed.
• flake8 src tests scripts --count --select=E9,F63,F7,F82 --show-source --statistics — passed.
• Python 3.12 compilation of the changed AI source and tests, git diff --check, and staged secret scan — passed.
• An offline harness loading the real AI package and using the real SDK with a mocked HTTP transport passed 16 requests with the repository-pinned SDK 0.117.0, and separately with SDK 1.4.0. Coverage includes default/explicit models, sync/stream serialization, default thinking/effort omission, image/video blocks, tools, and reasoning history across turns.
• Added tests/test_ai_minimax.py for transport and selector regression coverage. Running it alongside test_ai_ollama.py and test_ai_cancellation.py could not collect because the SDK-only environment lacks the GUI dependency hintedtext. The unchanged base commit 2018f5f76c43f948a53013b07184e375d85bb97d reproduces the same collection error. Full application dependency setup was paused when available disk fell below the shared environment's required 100 GiB threshold. This limitation applies to the local full application environment.
• Final-head CI at 32d0f7d33ea33b0b449083eb6fa898a4ac14a040 passes all nine platform test jobs, lint, documentation validation, DeepSource, and coverage checks. The Ubuntu/Python 3.13 log confirms all 23 new regression cases ran and the full suite reports 480 passed, 6 skipped, 1 deselected. CI run.
• The remaining red Vercel status says "Authorization required to deploy"; it is an external deployment gate.

No paid inference was performed. The integration retains its existing global Messages endpoint; no additional regional or gateway support is claimed. Explicit legacy/custom IDs are preserved, but their availability remains an API decision.

[vercel]

@octo-patch is attempting to deploy a commit to the Billy's projects Team on Vercel.

A member of the Team first needs to authorize it.

[codecov]

Codecov Report

❌ Patch coverage is 97.61905% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 38.48%. Comparing base (31dcd15) to head (32d0f7d).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/biscuit/views/ai/ai.py	83.33%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #825      +/-   ##
==========================================
+ Coverage   37.75%   38.48%   +0.72%     
==========================================
  Files         348      348              
  Lines       21946    21954       +8     
==========================================
+ Hits         8286     8448     +162     
+ Misses      13660    13506     -154     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.