 |
The DeepseekMath-V2 model achieves impressive performance in solving math problems. Photo: Decoder . |
Deepseek has just announced its DeepseekMath-V2 model, which has achieved gold medal-level results at a series of prestigious math competitions. This move puts the Chinese company in direct competition with leading US AI labs, which are leading in language model reasoning.
According to the technical documentation, DeepseekMath-V2 achieved gold medal-level results at the 2025 International Mathematical Olympiad (IMO) and the 2024 China CMO. In the Putnam exam, the model scored 118/120 points, far exceeding the 90-point mark of the highest-scoring human contestant in history. These results are a testament to its ability to handle complex problems.
The Chinese company explained that the limitation of previous AI models was that while they could give correct answers, they could not present valid solutions. To overcome this, DeepseekMath-V2 uses a multi-stage process that includes evaluating correctness, checking for counterarguments, and presenting the final result. This way of working allows the system to self-review and refine solutions in real time, avoiding the situation of correct conclusions but incorrect reasoning.
Notably, Deepseek's documentation makes no mention of the model using external tools like computers or interpreters. The test setup shows that all results are generated entirely in natural language.
Deepseek emphasizes that the performance gains come from the ability to self-criticize and self-correct solutions, not relying on specialized mathematical software. For difficult problems, the system scales up the computation by generating multiple solutions in parallel, thereby increasing the reliability of the final result.
The arrival of DeepseekMath-V2 comes shortly after OpenAI and Google DeepMind announced that their unpublished models also won gold medals at IMO, a feat that was previously beyond the capabilities of large language models. However, information about these systems is currently very limited, except that an OpenAI researcher said that a more mathematically powerful version will be available in the coming months.
In contrast to the secrecy of American companies, Deepseek has chosen to publish details of its methods and technical architecture. This approach allows the company to demonstrate transparency while also creating competitive pressure on the American AI ecosystem. According to The Economist , some American startups have considered switching to the low-cost open-source model from China to save costs.
Source: https://znews.vn/deepseek-pha-ky-luc-post1606990.html
Comment (0)