SeaEval - Multilingual Reasoning - MMLU (Zero-Shot): leaderboard
SeaEval zero-shot MMLU evaluation measuring multilingual subject knowledge and exam-style reasoning accuracy.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 81.29 | #454 |
| 2 | Llama 3.1 70B Instruct | 80.59 | #561 |
| 3 | Qwen 2.5 32B Instruct | 79.96 | #498 |
| 4 | Qwen 2 72B Instruct | 79.23 | #576 |
| 5 | Llama 3 70B Instruct | 76.5 | #634 |
| 6 | Qwen 2.5 14B Instruct | 75.42 | #647 |
| 7 | Gemma 2 9B (IT) | 71 | #788 |
| 8 | gemma2-9B-cpt-sea-lionv3-instruct | 70.79 | #705 |
| 9 | Qwen 2.5 7B Instruct | 69.35 | #862 |
| 10 | Qwen 2 7B Instruct | 67.25 | #991 |
| 11 | SeaLLMs-v3-7B-Chat | 66.7 | #915 |
| 12 | Llama 3.1 8B Instruct | 63.88 | #1035 |
| 13 | Sailor2-8B-Chat | 62.02 | #782 |
| 14 | Llama 3 8B Cpt Sea Lionv2 Instruct | 61.31 | #1188 |
| 15 | Qwen 2.5 3B Instruct | 61.19 | #1162 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seaeval-multilingual-reasoning-mmlu-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.