Question: Why is gpt-4.1-mini still the default INFERENCE_TEXT_MODEL when gpt-5-mini is available?

[adripo]

I noticed that the default INFERENCE_TEXT_MODEL is currently set to gpt-4.1-mini.

I saw that gpt-5-mini is now available. Based on the details, it seems to be cheaper and more intelligent compared to gpt-4.1-mini.

Is there a specific reason why we haven't updated the default to gpt-5-mini yet? Are there any known issues or blockers with using gpt-5-mini as the default model?

Thank you!

[MohamedBassem]

Hey, yes, you can use gpt-5-mini and in the cloud offering, I already use the gpt-5 family of models. Two caveats to be aware of:

1. The new models have higher latency (I assume because of reasoning), so inference will be slower compared to the 4 family of models.
2. You'll need to set INFERENCE_USE_MAX_COMPLETION_TOKENS=true for the 5 family as it's a requirement from openAI.

I'm planning to land some per-model evals (#2684) and soon after will change the default model. But for now, you can use it yourself.