SeaEval - Multilingual Reasoning - ZBench (Zero-Shot): leaderboard
SeaEval zero-shot ZBench evaluation measuring multilingual subject knowledge and exam-style reasoning accuracy.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 7B Instruct | 72.73 | #991 |
| 2 | Qwen 2.5 72B Instruct | 69.7 | #454 |
| 3 | Qwen 2.5 7B Instruct | 66.67 | #862 |
| 4 | Qwen 2.5 14B Instruct | 66.67 | #647 |
| 5 | Qwen 2.5 32B Instruct | 60.61 | #498 |
| 6 | Qwen 2 72B Instruct | 57.58 | #576 |
| 7 | Qwen 2.5 3B Instruct | 57.58 | #1162 |
| 8 | Llama 3.1 70B Cpt Sea Lionv3 Instruct | 57.58 | #548 |
| 9 | SeaLLMs-v3-7B-Chat | 54.55 | #915 |
| 10 | Llama 3 70B Instruct | 51.52 | #634 |
| 11 | Sailor2-8B-Chat | 51.52 | #782 |
| 12 | Llama 3.1 70B Instruct | 48.48 | #561 |
| 13 | Gemma 2 9B (IT) | 48.48 | #788 |
| 14 | Qwen 2.5 1.5B Instruct | 42.42 | #1407 |
| 15 | gemma2-9B-cpt-sea-lionv3-instruct | 42.42 | #705 |
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-zbench-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.