LEXam: leaderboard
7,537 questions from 340 University of Zurich law exams in English and German: 2,841 open-ended questions graded by an LLM judge plus 4,696 multiple-choice items with 4 to 32 options; 2025, ICLR 2026.
Metric: Multiple-Choice Accuracy (%). Source: github.com. Source: lexam-benchmark.github.io. Saturation forecast: Around January 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 62.65 | #98 |
| 2 | Claude Sonnet 4.5 | 58.01 | #142 |
| 3 | Claude 3.7 Sonnet | 57.23 | #249 |
| 4 | Gemini 2.5 Pro | 55.72 | #152 |
| 5 | GPT-5 Mini | 54.82 | #186 |
| 6 | GPT-4.1 | 54.4 | #248 |
| 7 | GPT-4o | 53.13 | #343 |
| 8 | DeepSeek V3.2 Exp | 53.07 | #234 |
| 9 | DeepSeek R1 | 52.41 | #251 |
| 10 | Llama 4 Maverick | 49.1 | #464 |
| 11 | GPT-4.1 Mini | 48.49 | #355 |
| 12 | Qwen 3 235B A22B | 48.19 | #313 |
| 13 | QwQ-32B | 47.83 | #421 |
| 14 | GPT-OSS-120B | 47.71 | #341 |
| 15 | GPT-5 Nano | 47.11 | #425 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=lexam · How It Works · Data refreshed daily, snapshot 2026-10-20.