LegalBench-RU - Reasoning: leaderboard
The reasoning track of LegalBench-RU — 348 cases requiring inference over facts, interpretation, arithmetic and Supreme/Constitutional Court positions, including resistance to a superseded edition of a norm; accuracy %.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 | 91 | #205 |
| 2 | Qwen 3 235B A22B | 85 | #313 |
| 3 | GPT-OSS-120B | 76 | #341 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=legalbench-ru-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.