Editorial illustration for DeepSeekMath-V2 Conquers Math Olympiads, Wins Gold and Top Honors in China
DeepSeekMath-V2 Conquers Math Olympiads with AI Gold Medal
DeepSeekMath-V2 Wins Gold at IMO 2025, Tops China Math Olympiad
DeepSeekMath-V2 just aced the hardest math tests on Earth. Gold at the International Mathematical Olympiad 2025, a top score on the China Mathematical Olympiad, a 118 out of 120 on the Putnam. It beat the best human Putnam score by a wide margin.
These aren't just wins. They are demolitions.
Yet the company behind it seems almost bored by the medals. The real problem, they argue, is that current AI is good at guessing answers and bad at thinking. Getting the final number on a benchmark is one thing.
Constructing a sound, step-by-step proof is something else entirely. So they built a model to critique its own work.
Besides the IMO 2025 competition, DeepSeekMath-V2 also achieved top-tier performance on China's toughest national competition, the China Mathematical Olympiad (CMO), and posted near-perfect results on the undergraduate Putnam exam. "On Putnam 2024, the preeminent undergraduate mathematics competition, our model solved 11 of 12 problems completely and the remaining problem with minor errors, scoring 118/120 and surpassing the highest human score of 90," stated DeepSeek. DeepSeek argues that recent AI models excel at getting the right answers (in math benchmarks like AIME and HMMT) but often lack sound reasoning. "Many mathematical tasks like theorem proving require rigorous step-by-step derivation rather than numerical answers, making final answer rewards inapplicable." To address this, DeepSeek emphasises the need for models that can judge and refine their own reasoning.
The scores are a fact. The reasoning is a philosophy. Everyone is chasing models that output correct answers.
DeepSeek is quietly building one that understands, and can defend, why its answer is correct. That's a different kind of machine. The gold medal proves it can compete.
The method suggests it might eventually learn.
Related Reading
Open Source Trump cracks down on Anthropic after Amazon tip; staff largely foreign
Open Source SDOF Adds Two Defensive Layers via Intent Router and StateAwareDisp
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
From Archives SAM3 uses concept segmentation to locate any object described in images or video
From Archives Startup Aims to Reinvent AI Interconnect, Tackling the Traffic Jam
Common Questions Answered
How did DeepSeekMath-V2 perform at the International Mathematical Olympiad (IMO) 2025?
DeepSeekMath-V2 achieved a gold medal at the IMO 2025, demonstrating exceptional mathematical problem-solving capabilities. This performance represents a significant breakthrough in AI's ability to tackle complex mathematical challenges traditionally reserved for top human mathematicians.
What remarkable score did DeepSeekMath-V2 achieve on the Putnam exam?
DeepSeekMath-V2 scored an impressive 118/120 on the Putnam exam, surpassing the highest human score of 90. The AI solved 11 of 12 problems completely and had only minor errors on the remaining problem, showcasing its near-superhuman mathematical reasoning abilities.
What makes DeepSeekMath-V2's performance in mathematical competitions so significant?
DeepSeekMath-V2 has broken new ground by not only competing in elite mathematical competitions but dominating them, including top performances at the International Mathematical Olympiad and China's Mathematical Olympiad. This achievement represents a major milestone in artificial intelligence's capability to solve complex mathematical problems at a level previously thought impossible for machines.
Further Reading
- How Does DeepSeekMath-V2 Achieve Self-Verifying Mathematical Reasoning — Dev.to
- DeepSeek has launched the DeepSeekMath-V2 model, focusing on self-verification training framework — Futunn News
- Advanced version of Gemini with Deep Think officially achieves gold medal standard at the International Mathematical Olympiad — Google DeepMind Blog
- Winning Gold at IMO 2025 with a Model-Agnostic Approach — arXiv
