On November 27th, DeepSeek unveiled a brand - new mathematical reasoning model, DeepSeekMath - V2, on the renowned HuggingFace platform. This innovative model is constructed on the foundation of DeepSeek - V3.2 - Exp - Base and adopts a self - verifiable training framework. In the process, it leverages an LLM (Large Language Model) verifier to automatically scrutinize mathematical proofs. Moreover, it enhances its performance by tackling challenging samples.

The model has achieved remarkable success, clinching gold medals in both the IMO (International Mathematical Olympiad) 2025 and the CMO (Chinese Mathematical Olympiad) 2024. Additionally, it scored an impressive 118 out of 120 in the Putnam 2024 competition. To promote the development of the field, DeepSeek has generously made the model's code and weights open - source, allowing researchers and developers worldwide to access and build upon this cutting - edge technology.