On April 30, the Chinese startup uploaded its new Prover-V2 model to Hugging Face, the largest platform for open source AI models, without even bothering to announce it on social media. Mathematical genius in the world of algorithms DeepSeek's Prover series are highly specialized models designed to solve mathematical problems. Interestingly, the company has not yet provided any details about the new model on its Hugging Face page. The files that have been viewed suggest that the Prover-V2 is based on the DeepSeek V3 model with an impressive 671 billion parameters. V3 uses a ”mixture of experts” architecture, an approach that enables cost-effective training and operation. This architecture is a smart move to maximize performance at minimal cost. Imagine that instead of hiring one expensive super-expert, you hire a team of highly specialized experts who only work when you need them - that's the savings! Competition between technology giants The launch of Prover-V2 comes just after Alibaba released its Qwen3. The e-commerce giant claimed, citing tests, that their newest model outperforms DeepSeek-R1 and OpenAI's o1 reasoning models. Prover-V2 is an update to the previous version, Prover-V1.5, which debuted in August, four months before DeepSeek wowed the world with its V3 model. The company claimed that V3 was developed at a fraction of the cost and energy that Western competitors spend on training advanced AI models. ”Our team has achieved an efficiency breakthrough,” DeepSeek executives might be heard saying. - We've proven that innovation doesn't necessarily require giant data centers and budgets the size of a small country's GDP.” Math as the key to advanced AI In a technical report on Prover-V1.5, DeepSeek noted that work on pre-training a specialized model has improved the capabilities of the underlying model in formal theorem proving and mathematical reasoning. The development of a math-oriented model that can augment the math skills of the underlying multi-purpose model has fueled speculation that DeepSeek will soon launch additional products. While DeepSeek doesn't officially share a timeline or progress on new models, the company regularly publishes its latest research results, including updates to the Prover model. Last month, DeepSeek also released an update to its base V3 model, which, according to the company's website, features improved reasoning capabilities, optimized programming, and updated Chinese writing skills. DeepSeek is expected to unveil a new R2 reasoning model soon, amid growing interest in this area of the AI industry.
