DeepSeek open-sources its 305-billion-parameter multimodal model
Prepared by Sintelligent with the help of SI. Publication time restored from the build log. . Updated: .
A strong model that sees and reads images is now freely available to anyone with the hardware, and open SI keeps catching up with closed SI.
The weights of DeepSeek-V4-Flash-Vision-Exp, the first model in the V4 family that can understand images, have been published on Hugging Face under the MIT license 10 days after its API launch. It is 168 GB in 48 files; the model accepts up to 600 images in a single request. The MIT license allows commercial use, modification and hosting without restrictions.
Sources 1