SeaEval - Cross-Lingual Consistency - Cross-XQuAD (Zero-Shot): leaderboard
SeaEval zero-shot Cross-XQuAD evaluation measuring answer accuracy and cross-lingual consistency across English, Chinese, Spanish, Vietnamese, Indonesian, Malay, and Filipino.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 96.83 | #454 |
| 2 | Llama 3.1 70B Instruct | 96.16 | #561 |
| 3 | Qwen 2 72B Instruct | 96.13 | #576 |
| 4 | Llama 3 70B Instruct | 95.92 | #634 |
| 5 | Qwen 2.5 14B Instruct | 95.82 | #647 |
| 6 | gemma2-9B-cpt-sea-lionv3-instruct | 95.74 | #705 |
| 7 | Gemma 2 9B (IT) | 95.67 | #788 |
| 8 | Qwen 2.5 7B Instruct | 94.6 | #862 |
| 9 | Qwen 2 7B Instruct | 94.18 | #991 |
| 10 | SeaLLMs-v3-7B-Chat | 94.03 | #915 |
| 11 | Llama 3.1 8B Instruct | 93.8 | #1035 |
| 12 | Qwen 2.5 3B Instruct | 93.78 | #1162 |
| 13 | Llama 3 8B Cpt Sea Lionv2 Instruct | 93.66 | #1188 |
| 14 | Llama 3 8B Cpt Sea Lionv2.1 Instruct | 93.61 | #1024 |
| 15 | Llama 3 8B Instruct | 92.1 | #1129 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seaeval-cross-lingual-consistency-cross-xquad-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.