2 ms·

v0.125.0 finally broke open models including their own gpt-oss over llama.cpp or vllm. I don't think they will fix it.