reasoning benchmark · included in ECI
HellaSwag
HellaSwag benchmark (score column: Overall accuracy).
Results
35
Random baseline
25.0%
Score ceiling
100%
Released
19 May 2019
Score by model release date
Best score by organisation
Leaderboard
| # | Model | Score | Relative | Setting | Released | Run | Evidence |
|---|---|---|---|---|---|---|---|
| 1 | Llama 3.1-405B Open source Meta AI | 89.2% | — | 23 Jul 2024 | — | Source ↗ | |
| 2 | Falcon-180B Open source Technology Innovation Institute | 89.0% | — | 6 Sep 2023 | — | Source ↗ | |
| 3 | DeepSeek V3 Open source DeepSeek | 88.9% | — | 26 Dec 2024 | — | Source ↗ | |
| 4 | DeepSeek-V2 (MoE-236B, May 2024) Open source DeepSeek | 87.1% | — | 7 May 2024 | — | Source ↗ | |
| 5 | Mixtral 8x7B Open source Mistral AI | 86.7% | — | 11 Dec 2023 | — | Source ↗ | |
| 6 | Llama 2-70B Open source Meta AI | 85.3% | — | 18 Jul 2023 | — | Source ↗ | |
| 7 | Falcon-40B Open source Technology Innovation Institute | 85.3% | — | 25 May 2023 | — | Source ↗ | |
| 8 | Qwen2.5-72B Open source Alibaba | 84.8% | — | 19 Sep 2024 | — | Source ↗ | |
| 9 | LLaMA-65B Open source Meta AI | 84.2% | — | 24 Feb 2023 | — | Source ↗ | |
| 10 | Stable Beluga 2 Open source Stability AI | 84.1% | — | 20 Jul 2023 | — | Source ↗ | |
| 11 | Qwen2.5-Coder-32B Open source Alibaba | 83.0% | — | 18 Sep 2024 | — | Source ↗ | |
| 12 | Falcon 2 11B Open source Technology Innovation Institute | 82.9% | — | 9 May 2024 | — | Source ↗ | |
| 13 | LLaMA-33B Open source Meta AI | 82.8% | — | 24 Feb 2023 | — | Source ↗ | |
| 14 | phi-3-medium 14B Open source Microsoft | 82.4% | — | 23 Apr 2024 | — | Source ↗ | |
| 15 | Gemma 7B Open source Google DeepMind | 82.2% | — | 21 Feb 2024 | — | Source ↗ | |
| 16 | Mistral 7B v0.1 Open source Mistral AI | 81.0% | — | 27 Sep 2023 | — | Source ↗ | |
| 17 | Llama 2-13B Open source Meta AI | 80.7% | — | 18 Jul 2023 | — | Source ↗ | |
| 18 | LLaMA-13B Open source Meta AI | 79.2% | — | 24 Feb 2023 | — | Source ↗ | |
| 19 | Falcon-7B Open source Technology Innovation Institute | 78.1% | — | 24 Apr 2023 | — | Source ↗ | |
| 20 | Llama 2-7B Open source Meta AI | 77.2% | — | 18 Jul 2023 | — | Source ↗ | |
| 21 | phi-3-small 7.4B Open source Microsoft | 77.0% | — | 23 Apr 2024 | — | Source ↗ | |
| 22 | Qwen2.5-Coder (7B) Open source Alibaba | 76.8% | — | 18 Sep 2024 | — | Source ↗ | |
| 23 | phi-3-mini 3.8B Open source Microsoft | 76.7% | — | 23 Apr 2024 | — | Source ↗ | |
| 24 | MPT-7B Open source MosaicML | 76.4% | — | 5 May 2023 | — | Source ↗ | |
| 25 | LLaMA-7B Open source Meta AI | 76.2% | — | 24 Feb 2023 | — | Source ↗ | |
| 26 | Yi 6B Open source 01.AI | 74.4% | — | 22 Nov 2023 | — | Source ↗ | |
| 27 | XGen-7B Open source Salesforce | 74.2% | — | 27 Jun 2023 | — | Source ↗ | |
| 28 | Gemma 2B Open source Google DeepMind | 71.4% | — | 21 Feb 2024 | — | Source ↗ | |
| 29 | Baichuan2-13B Open source Baichuan | 70.8% | — | 6 Sep 2023 | — | Source ↗ | |
| 30 | Dolly 2.0-12b Open source Databricks | 70.8% | — | 11 Apr 2023 | — | Source ↗ | |
| 31 | Baichuan 2-7B Open source Baichuan | 68.0% | — | 20 Sep 2023 | — | Source ↗ | |
| 32 | Qwen2.5-Coder (1.5B) Open source Alibaba | 61.8% | — | 18 Sep 2024 | — | Source ↗ | |
| 33 | Cerebras-GPT-13B Open source Cerebras Systems | 59.4% | — | 20 Mar 2023 | — | Source ↗ | |
| 34 | Phi-2 Open source Microsoft | 53.6% | — | 12 Dec 2023 | — | Source ↗ | |
| 35 | Phi-1.5 Open source Microsoft | 47.6% | 5 | 11 Sep 2023 | — | Source ↗ |
Caveats: settings (reasoning effort, agent scaffold) differ between rows and materially affect scores; the best reported setting per model is shown. Source: Epoch AI, CC BY 4.0.