91.2%
#1
MMLU-Pro accuracy of Gemini-3.1-Pro, the highest of the top 15 in early 2026; Gemini-3-Pro 90.1%, GPT-o1 89.3%
What the chart shows
MMLU-Pro tests knowledge and reasoning across subjects with more than 12,000 multiple-choice questions, each with ten options. In early 2026 Gemini-3.1-Pro led with 91.2% accuracy, ahead of Gemini-3-Pro (90.1%) and GPT-o1 (89.3%).
All of the top 15 models scored at least 87.0%, so 4.2 percentage points separate first from fifteenth. The report notes that accuracy on MMLU-Pro is typically 16–33% lower than on the original MMLU, which leaves more room to tell top models apart.
Key numbers
| Measure | Value |
|---|---|
| Gemini-3.1-Pro | 91.16% |
| Gemini-3-Pro (11/25) | 90.1% |
| GPT-o1 | 89.3% |
| Claude-4.6-Opus (Thinking) | 89.1% |
| Grok-4 (15th) | 87.0% |
How to read it
- Snapshot of the MMLU-Pro leaderboard (TIGER-Lab) in early 2026, top 15 models only.
- Accuracy is the share of questions answered correctly; guessing at random would score 10%.
- The report calls the second model Gemini-3-Pro (Thinking); the data labels it Gemini-3-Pro (11/25).
Source: MMLU-Pro Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.2.1: MMLU-Pro: overall 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