Skip to content

fix: correct provider MPS rows - #5

Merged
guillaumegay13 merged 1 commit into
mainfrom
codex/openai-anthropic-mps-fixes
May 21, 2026
Merged

fix: correct provider MPS rows#5
guillaumegay13 merged 1 commit into
mainfrom
codex/openai-anthropic-mps-fixes

Conversation

@guillaumegay13

@guillaumegay13 guillaumegay13 commented May 21, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • correct OpenAI API-key MPS rows against the official Chat Completions schema and live behavior
  • correct Anthropic API-key MPS rows, including current dated Claude 4.x IDs and Opus 4.7 adaptive-thinking rules
  • correct DeepSeek API-key MPS rows so current DeepSeek models and legacy aliases expose the thinking toggle/effort without invalid sampling combinations

Official docs checked

  • OpenAI Chat Completions API: max_completion_tokens, temperature, top_p, and reasoning_effort model constraints
  • Anthropic Claude Opus 4.7 docs: adaptive thinking only, no fixed thinking budgets, no non-default sampling params
  • Anthropic effort docs/model overview: current Claude 4.x model IDs and effort behavior
  • DeepSeek Chat Completion + Thinking Mode docs: thinking.type, reasoning_effort, sampling behavior while thinking is enabled

Live smoke coverage

  • OpenAI API-key catalog re-smoked: accessible rows had no parameter-shape failures; chatgpt-4o-latest, o1-mini, and o1-preview returned 404 for this test key, so they are left as a separate coverage/pruning policy question
  • Anthropic targeted current rows: Opus 4.7 adaptive/effort rules pass; deprecated sampling and fixed-budget thinking fail as expected and are no longer exposed
  • DeepSeek targeted rows: deepseek-chat, deepseek-reasoner, deepseek-v4-flash, and deepseek-v4-pro pass for supported parameter combinations; sampling params are hidden while thinking is enabled
  • Post-patch dependency-aware smoke on changed Anthropic/DeepSeek rows: 118 live calls, 0 failures

Notes

  • This intentionally does not include the reusable live smoke harness from closed PR fix: harden OpenAI and Anthropic model params #4. That should remain a separate PR.
  • This PR removes DeepSeek rows rejected by the live API (deepseek-v3.1, deepseek-v3.2, deepseek-v4) and keeps the accepted aliases (deepseek-chat, deepseek-reasoner).

Local validation

  • npm run validate
  • npm test
  • npm run typecheck
  • npm run lint
  • npm run build
  • npx prettier --check .
  • git diff --check

@vercel

vercel Bot commented May 21, 2026

Copy link
Copy Markdown

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

Project Deployment Actions Updated (UTC)
modelparameters-dev Ready Ready Preview May 21, 2026 7:32am

Request Review

@guillaumegay13
guillaumegay13 force-pushed the codex/openai-anthropic-mps-fixes branch from d77d015 to e8552f1 Compare May 21, 2026 07:32
@guillaumegay13
guillaumegay13 merged commit b2ea063 into main May 21, 2026
2 checks passed
@guillaumegay13 guillaumegay13 changed the title fix: correct OpenAI and Anthropic MPS rows fix: correct provider MPS rows May 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant