Chinese AI laboratory DeepSeek has released its new mathematics-oriented AI model, Prover, as open source. This was reported by SCMP.
Prover-V2, available on Hugging Face without official announcements, is an update to Prover-V1.5, which debuted in August—several months before the high-profile release of R1. At that time, the company described the product as a model for formal theorem proving and mathematical reasoning.
The Prover series consists of specific LLM focused on mathematics. Details about the new tool have not been disclosed. Based on the uploaded files, it is built on DeepSeek V3, which has 671 billion parameters and uses a Mixture-of-Experts (MoE) architecture.
The release of DeepSeek came a day after Chinese tech giant Alibaba introduced a new family of AI models, Qwen3.
In March, DeepSeek released a major update for V3, enhancing reasoning, programming, and writing capabilities.
Previously, the company attracted attention from Chinese authorities. Employees are now working under stricter conditions.