Chinese AI lab DeepSeek has quietly upgraded Prover, its artificial intelligence model dedicated to tackling mathematical proofs and theorem-solving tasks. The enhanced version, Prover V2, was recently uploaded to the AI development platform Hugging Face, according to reports.
Prover V2 is built upon DeepSeek’s powerful general-purpose AI model, V3, boasting an impressive architecture with approximately 671 billion parameters. Parameters directly influence a model’s capacity to solve complex problems, and the sheer scale of this update marks a significant advancement over its predecessor. Additionally, the model employs a sophisticated mixture-of-experts (MoE) framework, which allows it to break down complex mathematical tasks into smaller sub-problems. These subtasks are then distributed among specialized modules, each tailored to efficiently handle their assigned computations.
The last major update to Prover occurred in August, when DeepSeek announced improvements aimed specifically at formal theorem proving and advanced mathematical reasoning.
DeepSeek has been active in refining its broader portfolio of AI models as well. Besides releasing an upgraded version of its general-purpose V3 platform recently, the firm is also anticipated to launch an enhanced edition of another specialized model, the R1 “reasoning” engine.
Earlier this year, reports had emerged indicating that DeepSeek was exploring outside funding for the first time since its founding. This latest technological advancement further underscores the company’s ambitions and growing influence within the global AI community.
