Korean Canonical Legal Benchmark: leaderboard
Korean legal benchmark with multiple-choice and essay-style reasoning tasks, including support for precedent-based answers.
Metric: Mean with supporting precedents (self-reported). Source: benchmarklist.com. Status: saturated. 31 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 82.1 |
| 2 | GPT-5 | 78.5 |
| 3 | Gemini 2.5 Flash | 77.4 |
| 4 | Claude Sonnet 4.5 | 76.85 |
| 5 | O3 | 76.4 |
| 6 | Claude 3.7 Sonnet | 72.8 |
| 7 | Claude Opus 4.1 | 71.8 |
| 8 | GPT-5 Mini | 71.15 |
| 9 | Qwen 3 235B A22B (Thinking) | 70.4 |
| 10 | GPT-4.1 | 68.55 |
| 11 | Claude Sonnet 4 | 66.75 |
| 12 | O4 Mini | 64.55 |
| 13 | Kimi K2 | 63.65 |
| 14 | DeepSeek V3 | 63.45 |
| 15 | DeepSeek R1 | 62.75 |
Interactive version: theaggregate.ai/benchmark?slug=korean-canonical-legal-benchmark · How It Works · Data refreshed daily, snapshot 2026-09-05.