Korean LLM Eval - HAE-RAE Bench: leaderboard
Metric: Accuracy (%, 0-shot). Source: github.com. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 (Medium) | 95.84 |
| 2 | GPT-5.6 Sol (Medium) | 94.64 |
| 3 | GPT-5.6 Luna (Medium) | 93.17 |
| 4 | DeepSeek V4 Flash (0731) (Max) | 92.98 |
| 5 | GPT-5.6 Terra (Medium) | 92.85 |
| 6 | Claude Sonnet 5 (Medium) | 92 |
| 7 | DeepSeek V4 Flash (0731) (High) | 91.87 |
| 8 | DeepSeek V4 Flash (0731) (Non-reasoning) | 84.92 |
Interactive version: theaggregate.ai/benchmark?slug=korean-llm-eval-hae-rae-bench · How It Works · Data refreshed daily, snapshot 2026-09-19.