Fix: Route gpt-5.1 models to Responses API

[cjgalione]

Problem

Customers using gpt-5.x-codex models (e.g., gpt-5.1-codex, gpt-5.2-codex, gpt-5.3-codex) with:

• The invoke() method for Braintrust prompts
• Online scorers

Were encountering errors because these models were being routed to the Chat Completions API (/v1/chat/completions) instead of the Responses API (/v1/responses).

Root Cause

The proxy routing logic only matched:

• gpt-5-pro
• gpt-5-codex
• o1-pro
• o3-pro

The pattern startsWith("gpt-5-codex") matched "gpt-5-codex" but NOT "gpt-5.1-codex", "gpt-5.2-codex", "gpt-5.3-codex", etc.

Solution

Added a flexible pattern in packages/proxy/src/proxy.ts:2092-2093 that matches any gpt-5 codex variant:

if (
  bodyData?.model?.startsWith("o1-pro") ||
  bodyData?.model?.startsWith("o3-pro") ||
  bodyData?.model?.startsWith("gpt-5-pro") ||
  (bodyData?.model?.startsWith("gpt-5") &&
    bodyData?.model?.includes("-codex"))  // ← NEW: Future-proof pattern
) {
  return fetchOpenAIResponsesTranslate({ ... });
}

This pattern matches models that:

1. Start with "gpt-5" AND
2. Contain "-codex"

This is a conservative but future-proof approach that only routes codex variants to the Responses API, while leaving other gpt-5 models on the Chat Completions API.

Models Fixed

✅ gpt-5-codex
✅ gpt-5.1-codex
✅ gpt-5.1-codex-mini
✅ gpt-5.2-codex
✅ gpt-5.3-codex
✅ Any future gpt-5.x-codex variants

Models Unchanged

Other gpt-5 models continue to use Chat Completions API:

• gpt-5-mini
• gpt-5-nano
• gpt-5.1
• gpt-5.1-2025-11-13
• gpt-5.1-chat-latest

Testing

• Updated test to validate flexible gpt-5.x-codex routing pattern
• All 19 tests in openai.test.ts pass
• Test validates that gpt-5.x-codex models with reasoning_effort: "minimal" are properly converted to reasoning.effort: "low" in the Responses API format

Impact

• Invoke() method: Now works with all current and future gpt-5.x-codex models
• Online scorers: Now work with all gpt-5.x-codex models
• Customer unblocked: Can use model-agnostic prompts with gpt-5 codex variants across multiple providers
• Future-proof: Automatically supports newly released codex models (gpt-5.2-codex, gpt-5.3-codex, etc.)

Files Changed

• packages/proxy/src/proxy.ts - Added flexible gpt-5.x-codex routing pattern
• packages/proxy/src/providers/openai.test.ts - Updated test to reflect flexible pattern

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
ai-proxy	Ready	Preview, Comment	Feb 5, 2026 10:51pm

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 921a22954f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[ankrgyl]

very tiny nit, i would probably test that it starts with "gpt" and ends with "codex" to be safe. but i dont think it matters.