91.9%
#1 of 9
IMO-ProofBench score of Aletheia, the highest of nine systems; Gemini 3 Deep Think scored 76.7%
What the chart shows
IMO-ProofBench asks AI systems to write full, rigorous proofs for 60 problems ranging from pre-olympiad to International Mathematical Olympiad level. Aletheia led at 91.9%, followed by Gemini 3 Deep Think (76.7%) and the Gemini Deep Think version that won gold at the 2025 olympiad (65.7%).
Scores then drop: GPT-5.2 Thinking (high) reached 35.7%, Gemini 3 Pro 30% and GPT-5.1 7.1%, a spread of 84.8 percentage points. The AI Index notes that most models are far better at reaching correct answers than at proving them.
Key numbers
| Measure | Value |
|---|---|
| Aletheia | 91.9% |
| Gemini 3 Deep Think | 76.7% |
| Gemini Deep Think (IMO Gold) | 65.7% |
| GPT-5.2 Thinking (high) | 35.7% |
| Gemini 3 Pro | 30.0% |
| GPT-5.1 | 7.1% |
How to read it
- The benchmark’s authors validated automated proof grading against human experts, but recommend human checks for high-stakes results.
- “Gemini Deep Think (IMO Gold)” is the version that solved five of six problems at the 2025 International Mathematical Olympiad.
- A breakdown in the IMO-Bench paper suggests some scores may partly reflect familiarity with past competition problems.
Source: IMO-ProofBench Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.7: IMO-ProofBench. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion