DeepSeek-V3-0324 released — improved reasoning and coding capabilities

AI Impact Summary

DeepSeek-V3-0324, a new model from DeepSeek AI, has been released on the Hub, representing a significant update to the R1 reasoning model. This release includes an MIT license, improved instruction following, and enhanced capabilities in code and math, as demonstrated by benchmark improvements across MMLU-Pro, GPQA, and AIME. Users can experiment with this model via Inference Providers like Fireworks, Hyperbolic, and Novita, or through client libraries like Hugging Face’s InferenceClient and the OpenAI client library.