Benchmarks / Reasoning
HellaSwag
Uses Epoch AI's external HellaSwag leaderboard. Random baseline is 25% for four-way multiple choice. Score: HellaSwag accuracy. Collected by Jake Boggs' benchmark dashboard.
HellaSwag
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-4 | high | 95.3% |
| 2 | Llama 3.1 405b | none | 89.2% |
| 3 | DeepSeek V3 | none | 88.9% |
| 4 | DeepSeek V2 | none | 87.1% |
| 5 | Mixtral 8x7B Instruct | none | 86.7% |
| 6 | GPT-3.5 | high | 85.5% |
| 7 | Llama 2 70b | none | 85.3% |
| 8 | Qwen 2.5 72B | none | 84.8% |
| 9 | Llama 6 5b | none | 84.2% |
| 10 | Qwen 2.5 Coder | none | 83.0% |
| 11 | Llama 3 3b | – | 82.8% |
| 12 | Mistral | none | 81.0% |
| 13 | Llama 2 13b | none | 80.7% |
| 14 | Llama 1 3b | – | 79.2% |
| 15 | GPT-3 | medium | 77.5% |
| 16 | Llama 2 7b | none | 77.2% |
| 17 | GPT-2 | – | 40.0% |