DeepSeek open-sourced DeepSeek-V4.1-Flash-1017 on October 17, 2026, under the MIT License. The weights are available on Hugging Face and ModelScope. The model permits commercial use, and the vLLM and SGLang communities already provide support. We request that it be added to the NVIDIA NIM API.
It may come in next 1-2 week cause dev team need to test its compatibility and performance
OMG, their old version hasn’t been working for days now, and you’re asking for a new one)
it worked fine for me, fast and good
For the first time in a week, I see that the model is responding, though very slowly.
Me is normal:
- average tps: ~40 T/S
- Average time to first token: 8-40 sec
- Highest tps: ~60-70 t/s
Maybe it depends on the location? I’m in Germany, and it’s very slow.
Maybe but now deepseek-v4.1 is heavily limited
always 429 errors but others work fine