Magis-Bench: leaderboard
Existing benchmarks for legal AI focus primarily on tasks where LLMs must produce legal arguments or documents, yet the capacity to judge such arguments -- weighing competing claims, applying doctrine to facts, and rendering reasoned decisions -- is arguably as fundamental to a well-functioning legal system as advocacy itself. We introduce Magis-Bench, a benchmark for evaluating LLMs on ma
Metric: AVG (self-reported). Source: benchmarklist.com. Saturation forecast: Around March 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 6.97 | #66 |
| 2 | Gemini 3 Flash (Preview) | 6.67 | #82 |
| 3 | Claude Opus 4.5 | 6.46 | #83 |
| 4 | GPT-5.1 | 6.23 | #135 |
| 5 | Gemini 2.5 Pro | 6.18 | #152 |
| 6 | Claude Sonnet 4.5 | 5.55 | #142 |
| 7 | GPT-4.1 | 5.2 | #248 |
| 8 | DeepSeek V3.2 | 4.34 | #205 |
| 9 | Kimi K2.5 | 4.17 | #144 |
| 10 | Kimi K2 (Thinking) | 4.14 | #243 (Kimi K2) |
| 11 | GPT-5 Mini | 4.08 | #186 |
| 12 | Qwen 3 235B A22B (Thinking) | 4 | #313 (Qwen 3 235B A22B) |
| 13 | GPT-4.1 Mini | 3.84 | #355 |
| 14 | Qwen 2.5 72B Instruct | 2.62 | #454 |
| 15 | Qwen 2.5 14B Instruct | 2.04 | #647 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=magis-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.