Chinese AI lab DeepSeek has secretly introduced a significant update to Prover, its advanced artificial intelligence (AI)-based model.
Prover is particularly designed to solve math-related proofs and theorems.
DeepSeek updates Prover
The company uploaded the newest variant of Prover, V2, and a distilled version to the AI dev platform Hugging Face on Wednesday, April 30, 2025, as reported by South China Morning Post.
It seems to be built on top of the startup’s V3 model, which consists of 671 billion parameters and takes up a mixture-of-experts (MoE) architecture, the post further reads.
Parameters align with problem-solving capabilities, while MoE divides tasks into subtasks and assigns them to specialised “expert” components.
DeepSeek last received an update in August, explaining it at the time as a custom openly accessible AI-powered model for formal theorem proving and providing mathematical reasoning.
According to Reuters, In February the Chinese AI lab was claimed to be considering securing investment from outside for the first time.
The company recently launched an upgraded variant of V3, a general-purpose model, and is likely to update its R1 “reasoning” model in the near future.
In a technical report for Prover-V1.5, DeepSeek stated that it is currently working to pre-train the specialist model and advance its base model’s features in formal theorem proving and mathematical reasoning.
However, the company has yet to officially announce the timeline or progress for the latest models.
