Chinese AI start-up DeepSeek has generated renewed public interest with the unexpected release of its upgraded mathematics-focused model, Prover-V2, as anticipation builds for its next-generation R2 reasoning model. The newly open-sourced Prover-V2 boasts 671 billion parameters and is designed for mathematical proof tasks, marking an evolution from its predecessor Prover-V1.5 released in August. Though it is built on DeepSeek’s V3 foundational model, this release is seen as an incremental step rather than the breakthrough that the AI community has been eagerly awaiting.
Prover-V2 showcases DeepSeek’s capability to produce powerful, domain-specific models despite restricted access to Nvidia’s advanced AI chips. This latest release has earned praise from mathematics and AI enthusiasts online, with users describing it as impressive and hinting it may be a prelude to a broader leap in DeepSeek’s AI reasoning capabilities. The release prompted speculation across social media platforms like X and Reddit, with many interpreting it as groundwork for the company’s highly anticipated R2 model.
The announcement came just a day after Alibaba launched its Qwen3 series of large language models, which it claimed outperformed DeepSeek’s earlier R1 reasoning model across various benchmarks. DeepSeek-R1, which debuted in January, had previously made waves by delivering OpenAI’s o1-level performance at a fraction of the computational cost, making it a standout in the global AI scene. OpenAI has since responded with its newer o3 and o4-mini models, described as its most advanced to date.
While no official release date or specifications for R2 have been shared, the upgraded Prover model has fueled further speculation and excitement. Analysts and enthusiasts have interpreted recent DeepSeek activity, including its job postings, as signs that the firm is scaling up operations. The company is currently hiring for key positions, including product and design roles in Beijing and Hangzhou, and is also seeking a chief financial officer and chief operating officer.
Founded in 2023 by entrepreneur Liang Wenfeng as a spin-off from his hedge fund High-Flyer, DeepSeek remains tight-lipped about its roadmap. Still, the company’s research progress and minimal public commentary only seem to amplify curiosity about its next major release. With mounting global attention and rising Google Trends searches, expectations for R2 continue to grow as DeepSeek positions itself as a key player in China’s AI landscape.
READ MORE: