DeepSeek, a fast-growing AI research lab from China, has quietly rolled out a major update to its math-solving AI tool called Prover. The new version, known as Prover V2, is now available on Hugging Face, a popular platform for AI developers.
This update is based on DeepSeek’s powerful V3 model, which uses 671 billion parameters and a smart method called mixture-of-experts (MoE). This MoE approach lets the AI split problems into smaller parts and assign them to mini expert models, making it more efficient and accurate in solving complex math problems.
DeepSeek originally launched Prover to focus on formal theorem proving and deep mathematical reasoning. The last known update was in August, when the company highlighted its goal of creating an AI tool that can handle advanced mathematics, such as proofs and theorems.
The new Prover V2 is expected to take that mission further, building on the strength of the massive V3 foundation. This follows DeepSeek’s recent efforts to grow its influence in the AI space, including launching an upgraded V3 general-purpose model and planning an update for its R1 reasoning model.
Reports from earlier this year also revealed that DeepSeek might soon seek external funding for the first time, showing strong momentum and ambition in its AI development journey.
As DeepSeek quietly but consistently improves its tools, it positions itself as a serious competitor in the AI race, especially in the niche field of mathematical AI.
By
