← All benchmarks

International Physics Olympiad 2025 - theory

Science

Normalized partial-credit score on all three theory problems from the 2025 International Physics Olympiad. Google's February 2026 evaluation averages eight runs and uses Gemini judging against canonical solutions, with independent experts validating Deep Think's gold-medal-level performance. This is a mean solution score, not binary accuracy.

Top models (higher is better)