31.3%
▲ 31.3 pts since 2024
of FrontierMath Tier 4 problems solved at the first attempt by GPT-5.2 Pro, 2025
20242025
What the chart shows
FrontierMath is a set of original, very difficult maths problems from Epoch AI that can take expert mathematicians hours or days. On Tier 4, its hardest tier, the best result rose from 0% in 2024 (Claude 3.5 Sonnet, June 2024 version) to 31.3% in 2025 (GPT-5.2 Pro).
Pass@1 counts a problem as solved only if the model’s first answer is correct. Even the leading model still fails on about two in three of these problems.
Key numbers
| Measure | Value |
|---|---|
| 2024: Claude 3.5 Sonnet (Jun 2024) | 0.0% |
| 2025: GPT-5.2 Pro | 31.3% |
| Change, 2024 to 2025 | +31.3 percentage points |
| Problems still failed by the 2025 leader | 68.7% |
How to read it
- Each bar is the best result recorded in that year; the report names GPT-5.2 Pro (Web App) as the leader at the end of 2025.
- FrontierMath is designed to stay difficult, and Tier 4 is its hardest set.
Source: Epoch AI, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.4: FrontierMath Tier 4: pass@1 accuracy. 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