Back to BenchmarksVery Hard
HellaSwag
Commonsense NLI about grounded situations
Leaderboard
(17 models)| Rank | Model | Score | Stderr |
|---|---|---|---|
| 1 | GPT-4.1 | 95.30 | — |
| 2 | Llama 3.1 405B | 89.20 | — |
| 3 | falcon-180B | 89.00 | — |
| 4 | DeepSeek V3 | 88.90 | — |
| 5 | Mixtral-8x7B-v0.1 | 86.70 | — |
| 6 | Llama-2-70b-hf | 85.30 | — |
| 7 | Qwen 2.5 72B | 84.80 | — |
| 8 | Qwen2.5-Max | 83.00 | — |
| 9 | Phi-3-medium-128k-instruct | 82.40 | — |
| 10 | gemma-7b | 82.20 | — |
| 11 | Mistral-7B-v0.1 | 81.00 | — |
| 12 | Llama-2-7b | 80.70 | — |
| 13 | Phi-3-small-8k-instruct | 77.00 | — |
| 14 | Phi-3-mini-4k-instruct | 76.70 | — |
| 15 | Yi-6B | 76.40 | — |
| 16 | GPT-OSS 120B | 70.50 | — |
| 17 | Phi-4 | 53.60 | — |
Data source: Epoch AI, “Data on AI Benchmarking”. Published at epoch.ai
Licensed under CC-BY 4.0