SeaEval - Cultural Reasoning - SG-Eval v1 Cleaned (Zero-Shot): leaderboard
SeaEval zero-shot SG-Eval v1 Cleaned evaluation measuring culturally grounded reasoning and regional knowledge for Southeast Asian and comparison contexts.
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 | 72.06 | #454 |
| 2 | Qwen 2.5 14B Instruct | 69.12 | #647 |
| 3 | Qwen 2 72B Instruct | 67.65 | #576 |
| 4 | Llama 3.1 70B Instruct | 66.18 | #561 |
| 5 | Llama 3 70B Instruct | 66.18 | #634 |
| 6 | Llama 3 8B Cpt Sea Lionv2.1 Instruct | 66.18 | #1024 |
| 7 | Llama 3 8B Cpt Sea Lionv2 Instruct | 66.18 | #1188 |
| 8 | Qwen 2.5 32B Instruct | 64.71 | #498 |
| 9 | Qwen 2 7B Instruct | 63.24 | #991 |
| 10 | Gemma 2 9B (IT) | 60.29 | #788 |
| 11 | gemma2-9B-cpt-sea-lionv3-instruct | 60.29 | #705 |
| 12 | Qwen 2.5 7B Instruct | 58.82 | #862 |
| 13 | Llama 3 8B Instruct | 58.82 | #1129 |
| 14 | Qwen 2.5 3B Instruct | 58.82 | #1162 |
| 15 | SeaLLMs-v3-7B-Chat | 58.82 | #915 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seaeval-cultural-reasoning-sg-eval-v1-cleaned-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.