Fix gpt-5-chat incorrectly classified as reasoning model

[yashwantbezawada]

Summary

Fixes #9032 - Updates the reasoning model regex pattern to prevent gpt-5-chat from being incorrectly classified as a reasoning model.

Problem

The Azure gpt-5-chat model was being misclassified as a reasoning model, causing it to fail with inappropriate validation requirements:

• ValueError: reasoning models require passing temperature=1.0 and max_tokens >= 16000
• Even when complying, Azure API rejects the reasoning parameter with BadRequestError

Root Cause: The regex pattern lacked an end anchor ($), allowing any model starting with gpt-5- to match as a reasoning model.

Changes

Regex Pattern Update

Before:

r"^(?:o[1345]|gpt-5)(?:-(?:mini|nano))?"

After:

r"^(?:o[1345](?:-(?:mini|nano))?(?:-\d{4}-\d{2}-\d{2})?|gpt-5(?:-(?:mini|nano|pro))?)$"

Improvements:

• ✅ Added $ anchor for exact matching (prevents gpt-5-chat from matching)
• ✅ Added pro variant support (gpt-5-pro now recognized as reasoning model)
• ✅ Explicit date suffix support for o-series models (o1-2023-01-01)
• ✅ Separate handling for o-series vs gpt-5 models

Test Coverage

Added test cases to verify:

• ✅ gpt-5-pro correctly identified as reasoning model
• ✅ gpt-5-chat correctly identified as NON-reasoning model
• ✅ azure/gpt-5-chat correctly identified as NON-reasoning model

Verification

Tested regex pattern against all model variants:

Reasoning Models (should match):

• o1, o1-mini, o1-nano, o1-2023-01-01, o1-mini-2023-01-01
• o3, o3-mini, o3-mini-2023-01-01
• o4, o5
• gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro

Non-Reasoning Models (should NOT match):

• gpt-5-chat ✓
• gpt-4, gpt-4o
• o2, o6
• Other models

All tests pass correctly.

Impact

Users can now use gpt-5-chat with standard conversational parameters (temperature < 1.0, max_tokens < 16000) without triggering incorrect reasoning model validation.

[chenmoneygithub]

@yashwantbezawada Thanks for the PR!

Looks like it's already closed by #9033.