I'm exploring an alternative like using a native gpt client for mac and use chatgpt through the api instead. Matgpt supports gpt-4o mini as the default model, which delivers higher performance at a lower cost than gpt-3.5 turbo. Aug 5, 2025inference examples transformers you can use gpt-oss-120b and gpt-oss-20b with the transformers library.
If you use transformers' chat template, it will automatically apply the harmony.