Event Summary
In July 2025, Google DeepMind announced that an advanced version of Gemini with Deep Think had achieved a gold-medal standard at the International Mathematical Olympiad, scoring 35 out of 42 points. This marked the first time an AI system reached the highest level of the world's most prestigious mathematics competition for pre-university students — solving problems requiring deep mathematical reasoning, creativity, and multi-step logic. The achievement demonstrated that inference-time compute scaling could match the most challenging human intellectual benchmarks.
Context & Narrative
The International Mathematical Olympiad is the world's most challenging mathematics competition for high school students, requiring contestants to solve six proof-based problems across two days. In 2024, DeepMind's AlphaProof + AlphaGeometry 2 had achieved a silver medal (28/42 points). The 2025 breakthrough with Gemini Deep Think improved to 35/42 — a gold medal. Unlike AlphaProof which used formal mathematical verification, Gemini Deep Think used natural-language reasoning augmented by 'Deep Think,' a test-time compute technique that allowed the model to explore multiple solution paths, verify its reasoning, and refine its approach before finalizing an answer. This was significant for two reasons. First, it showed that general-purpose language models, not just specialized math systems, could reach the highest levels of mathematical reasoning. Second, the Deep Think approach — scaling inference-time compute rather than training larger models — validated the paradigm that o1 had introduced in September 2024. Google's approach was to apply Deep Think to an already capable Gemini model, dramatically improving its IMO performance without retraining. The result placed Gemini Deep Think among the top IMO gold medalists, outperforming most human contestants. The achievement was reported by Ars Technica, Nature, and major media outlets. It was quickly followed by similar results from OpenAI's o3 (which claimed gold without following formal IMO rules). The milestone had implications beyond mathematics: it suggested that with sufficient inference-time compute, AI could match human experts across STEM fields requiring rigorous reasoning.
Key Findings
-
Fact Grade A
Google DeepMind's Gemini with Deep Think achieved a gold-medal score of 35/42 at the International Mathematical Olympiad in July 2025.
Sources [1] -
Impact Grade A
AI reached gold-medal level at the International Mathematical Olympiad — the world's most challenging pre-university math competition. Scored 35/42, outperforming most human contestants. Demonstrated general-purpose LLMs could match top human mathematical reasoning.
Sources [1]
Impact Assessment
-
Capability Leap +3 · Long-term
AI reached gold-medal level at the International Mathematical Olympiad — the world's most challenging pre-university math competition. Scored 35/42, outperforming most human contestants. Demonstrated general-purpose LLMs could match top human mathematical reasoning.
Affected Groups: mathematicians, AI researchers, educators, students
-
Paradigm Shift +2 · Long-term
Validated inference-time compute scaling as a path to expert-level reasoning. Showed that spending more compute at inference time (Deep Think) could match training larger models — reinforcing the reasoning model paradigm introduced by o1.
Affected Groups: AI researchers, ML engineers, AI companies
Consensus & Sources
-
1
Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad.Reference Evidence Citation logged Live source
-
2
Gemini Deep Think earns gold at International Math Olympiad.News Report Citation logged Live source