The docs say haiku 4.5 or gpt 5.4-mini Run Modes | Cursor Docs.
If that’s really the case, why not use newer, cheaper and more intelligent models like gpt 5.6 luna instead?
Hey @Yoan_Babulkov, thanks for the question!
The choice isn’t only about raw capability: the classifier is a latency-sensitive, prompt-calibrated path, so models are picked based on how well they do at this specific task. We regularly evaluate newer models against it, and the docs will be updated whenever the lineup changes!
@sinar_pandora unrelated. Feel free to start a new thread!