SeaEval - Fundamental NLP Tasks - MNLI (Zero-Shot): leaderboard
SeaEval zero-shot MNLI evaluation measuring core NLP understanding such as inference, acceptability, paraphrase, or reading-comprehension accuracy.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 32B Instruct | 87.15 | #498 |
| 2 | Qwen 2.5 72B Instruct | 84.45 | #454 |
| 3 | Qwen 2.5 14B Instruct | 81.8 | #647 |
| 4 | Qwen 2.5 7B Instruct | 81.05 | #862 |
| 5 | Qwen 2 72B Instruct | 79.25 | #576 |
| 6 | Qwen 2.5 3B Instruct | 74.65 | #1162 |
| 7 | gemma2-9B-cpt-sea-lionv3-instruct | 74.45 | #705 |
| 8 | Llama 3.1 70B Cpt Sea Lionv3 Instruct | 73.6 | #548 |
| 9 | Qwen 2 7B Instruct | 72.95 | #991 |
| 10 | Gemma 2 9B (IT) | 71.6 | #788 |
| 11 | Llama 3.1 70B Instruct | 70.15 | #561 |
| 12 | Llama 3 70B Instruct | 67.09 | #634 |
| 13 | Sailor2-8B-Chat | 66.4 | #782 |
| 14 | SeaLLMs-v3-7B-Chat | 65.3 | #915 |
| 15 | Gemma 2 2B (IT) | 61.85 | #1341 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seaeval-fundamental-nlp-tasks-mnli-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.