[octo-patch]

Reason: MiniMax-M3.1-Flash is the current first choice for ordinary MiniMax text chat, while existing explicit MiniMax-M3, legacy, saved, and custom model IDs remain accepted.

Summary

• Make MiniMax-M3.1-Flash the default and first ordinary model choice.
• Preserve MiniMax-M3 thinking controls, explicit legacy/custom model requests, chat history, and multimodal message content.
• Serialize Flash thinking effort as reasoning_effort for the OpenAI-compatible route and output_config.effort for the Anthropic-compatible route.
• Keep the existing global and China endpoint routing and provider registration unchanged.

Validation

• python3 -m py_compile passed for the changed processor and test files.
• git diff --check passed.
• python -m unittest llmstack.processors.providers.test_minimax -v: all 6 tests passed, including 44 processor HTTP captures and 4 existing endpoint requests. Tests cover both regional hosts, both compatibility protocols, default and saved/custom model IDs, thinking modes, all supported Flash effort values, history, and image/video payload preservation.
• Black 23.12.1 (--check --line-length 120), isort 5.13.2 (--check-only --profile black), and flake8 7.0.0 (--max-line-length=300 --extend-ignore=E203) passed for both changed files.
• Focused tests used Python 3.10, OpenAI 1.40.0, HTTPX 0.27.0, and Pydantic 2.8.2 with mocked HTTP transport and a credential-free environment. A regional package mirror resolved the initial download timeouts.

Limitations

• Live MiniMax inference was not run.
• Full application build and full locked dependency installation were not run because heavy setup is paused under the available disk threshold. The focused dependency subset used the repository's pinned SDK, HTTPX, Pydantic, and formatter versions; transitive dependencies were not a full lockfile reproduction.
• Native Responses routing was not changed; the existing Chat Completions compatibility paths remain the tested integration path for this provider.

[vercel]

@octo-patch is attempting to deploy a commit to the LLMStack Team on Vercel.

A member of the Team first needs to authorize it.