DeepSeek, a Chinese AI lab, has released an update to its DeepSeek V3 model, now available as DeepSeek-V3-0324 on Hugging Face. This update, which is licensed under MIT, is a significant enhancement to the base model, with the total size of the files amounting to 641 GB and 700 billion parameters.
The updated model has been improved through post-training on a new 32k GPU cluster, showing particular advancements in math and coding skills. Early user tests indicate that the model performs well, though official benchmarks are yet to be released. It is now available for testing on OpenRouter.
DeepSeek has also announced the closure of DeepThink as they shift their focus towards their website, app, and API. The release of DeepSeek-V3-0324 is seen as a precursor to the upcoming DeepSeek R2, with expectations of further advancements in AI capabilities. The model can be run on a 512GB M3 Ultra with 4-bit quantization, reducing its on-disk size to 352 GB.
