I can confirm it now works well. Currently running DeepSeek-R1 using the HuggingFace Inference API. There are intermittent 502 and 504 errors, but that’s the fault of HF I believe. Users should expect all free endpoints to be extremely busy for now lol.