DeepSeek updates V4.1 Flash: open weights and shorter answers
DeepSeek released the V4.1-Flash-1017 update with open weights under the MIT license, so it can run on your own server.
On October 17 DeepSeek released an updated version of its V4.1-Flash model, V4.1-Flash-1017. The company calls it a minor update: benchmark results improved and answers became shorter. The model is live in DeepSeek’s apps and API.
The weights are open on Hugging Face under the MIT license: the model can be downloaded, modified and run on your own server. It reads up to 1 million tokens at once and understands images.
Cheap, open models are making AI practical for small and mid-sized businesses too. Nova, the assistant on novacode.uz, runs on DeepSeek V4.1 Flash, and when data must not leave the company, such a model can be installed on the company’s own server.