How about Switch OpenAI-based tests to gpt-4.1-nano

[SongChiYoung]

Hi all,

Many of our test cases currently use gpt-4o or gpt-4o-mini. These models are powerful, but also relatively heavier in both latency and cost — especially for running full test suites repeatedly.

Right now, this is still manageable. But looking ahead, it’s easy to imagine scenarios where the cumulative cost or runtime becomes a bottleneck — especially as more tests, workflows, or contributors get added.

That’s why I’d like to suggest:
Let’s switch to gpt-4.1-nano for all OpenAI-based tests.

According to OpenAI’s official documentation, gpt-4.1-nano supports function calling, JSON mode, and logprobs — so structurally, it covers all the core API features we rely on in tests.

Since our tests are mostly structural (not evaluating model quality), the tradeoff seems minimal and the gain is long-term sustainability.

Open to feedback — happy to send a PR if folks agree.

[ekzhu]

I think it is a good idea.

[SongChiYoung]

Resolved it
At #6375