Reasoning benchmark # HellaSwag leaderboard > HellaSwag results for 29 AI models, led by GPT-4 at 95.3%. What the benchmark measures, who runs it, and a source for every score. - Canonical page: https://noometry.com/benchmarks/hellaswag - Last updated: 2026-10-10 - Title: HellaSwag Leaderboard (October 2026): Scores by Model As of October 2026, GPT-4 has the highest published HellaSwag score on Noometry at 95.3%, out of 29 models with results. Last verified October 10, 2026 ## About HellaSwag Commonsense sentence completion with adversarially filtered wrong endings. - **Category:** [Reasoning](https://noometry.com/best/reasoning) - **Introduced:** 2019 - **Format:** Multiple choice - **Unit:** Percent (random guessing ≈ 25%) - **Official site:** [rowanzellers.com](https://rowanzellers.com/hellaswag/) ## Top 15 models Top models on HellaSwag 1. GPT-4 95.3% 2. GPT-4 95.3% 3. Llama 3.1-405B 89.2% 4. Falcon-180B 89% 5. DeepSeek-V3 88.9% 6. DeepSeek-V2 (MoE-236B, May 2024) 87.1% 7. Mixtral 8x7B 86.7% 8. Llama 2-70B 85.3% 9. Falcon-40B 85.3% 10. Qwen2.5 72B Instruct 84.8% 11. Qwen2.5-Coder-32B 83% 12. Falcon 2 11B 82.9% 13. Nemotron-4 15B 82.4% 14. phi-3-medium 14B 82.4% 15. Gemma 7B 82.2% 16. 7580859095100 Sponsored placements are available on pages like this one. [Advertise on Noometry](https://noometry.com/advertise) ## All results HellaSwag results by model | # | Model | Provider | Score | Setting | Source | Date | | --- | --- | --- | --- | --- | --- | --- | | 1 | [GPT-4](https://noometry.com/models/gpt-4) | [OpenAI](https://noometry.com/providers/openai) | 95.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 2 | [GPT-4](https://noometry.com/models/gpt-4) | [OpenAI](https://noometry.com/providers/openai) | 95.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 3 | [Llama 3.1-405B](https://noometry.com/models/llama-3-1-405b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 89.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 4 | [Falcon-180B](https://noometry.com/models/falcon-180b) | [![](/logos/tii.svg) Technology Innovation Institute](https://noometry.com/providers/tii) | 89% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 5 | [DeepSeek-V3](https://noometry.com/models/deepseek-v3) | [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 88.9% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 6 | [DeepSeek-V2 (MoE-236B, May 2024)](https://noometry.com/models/deepseek-v2) | [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 87.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 7 | [Mixtral 8x7B](https://noometry.com/models/mixtral-8x7b) | [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 86.7% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 8 | [Llama 2-70B](https://noometry.com/models/llama-2-70b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 85.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 9 | [Falcon-40B](https://noometry.com/models/falcon-40b) | [![](/logos/tii.svg) Technology Innovation Institute](https://noometry.com/providers/tii) | 85.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 10 | [Qwen2.5 72B Instruct](https://noometry.com/models/qwen2-5-72b-instruct) | [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 84.8% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 11 | [Qwen2.5-Coder-32B](https://noometry.com/models/qwen2-5-coder-32b) | [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 83% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 12 | [Falcon 2 11B](https://noometry.com/models/falcon-2-11b) | [![](/logos/tii.svg) Technology Innovation Institute](https://noometry.com/providers/tii) | 82.9% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 13 | [Nemotron-4 15B](https://noometry.com/models/nemotron-4-15b) | [![](/logos/nvidia.svg) NVIDIA](https://noometry.com/providers/nvidia) | 82.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 14 | [phi-3-medium 14B](https://noometry.com/models/phi-3-medium-14b) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 82.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 15 | [Gemma 7B](https://noometry.com/models/gemma-7b) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 82.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 16 | [Mistral 7B](https://noometry.com/models/mistral-7b) | [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 81% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 17 | [Llama 2-13B](https://noometry.com/models/llama-2-13b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 80.7% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 18 | [Llama 13b](https://noometry.com/models/llama-13b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 79.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 19 | [Falcon-7B](https://noometry.com/models/falcon-7b) | [![](/logos/tii.svg) Technology Innovation Institute](https://noometry.com/providers/tii) | 78.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 20 | [Llama 2-7B](https://noometry.com/models/llama-2-7b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 77.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 21 | [Phi 3 Small 8k Instruct](https://noometry.com/models/phi-3-small-8k-instruct) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 77% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 22 | [Qwen2.5-Coder (1.5B)](https://noometry.com/models/qwen2-5-coder) | [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 76.8% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 23 | [Phi 3 Mini 4k Instruct](https://noometry.com/models/phi-3-mini-4k-instruct) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 76.7% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 24 | [Yi 6B](https://noometry.com/models/yi-6b) | [01.AI](https://noometry.com/providers/01-ai) | 74.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 25 | [INTELLECT-1](https://noometry.com/models/intellect-1) | [![](/logos/huggingface.svg) Hugging Face](https://noometry.com/providers/huggingface) | 71.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 26 | [Gemma 2B](https://noometry.com/models/gemma-2b) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 71.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 27 | [Dolly 2.0-12b](https://noometry.com/models/dolly-2-0-12b) | [![](/logos/databricks.svg) Databricks](https://noometry.com/providers/databricks) | 70.8% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 28 | [Phi-2](https://noometry.com/models/phi-2) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 53.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 29 | [Phi-1.5](https://noometry.com/models/phi-1-5) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 47.6% | 5 | [Epoch AI](https://epoch.ai/benchmarks) | | ## Compare the leaders - [GPT-4 vs GPT-4](https://noometry.com/compare/gpt-4-vs-gpt-4) - [GPT-4 vs Llama 3.1-405B](https://noometry.com/compare/gpt-4-vs-llama-3-1-405b) - [GPT-4 vs Falcon-180B](https://noometry.com/compare/falcon-180b-vs-gpt-4) - [GPT-4 vs DeepSeek-V3](https://noometry.com/compare/deepseek-v3-vs-gpt-4) - [GPT-4 vs Llama 3.1-405B](https://noometry.com/compare/gpt-4-vs-llama-3-1-405b) - [GPT-4 vs Falcon-180B](https://noometry.com/compare/falcon-180b-vs-gpt-4) ## Other reasoning benchmarks - [ARC-AGI-2](https://noometry.com/benchmarks/arc-agi-2) - [SimpleBench](https://noometry.com/benchmarks/simplebench) - [Kagi LLM Benchmark](https://noometry.com/benchmarks/kagi-reasoning) - [NYT Connections (extended)](https://noometry.com/benchmarks/nyt-connections) - [ARC-AGI-1](https://noometry.com/benchmarks/arc-agi-1) - [CritPt](https://noometry.com/benchmarks/critpt) - [Chess Puzzles](https://noometry.com/benchmarks/chess-puzzles) - [EnigmaEval](https://noometry.com/benchmarks/enigmaeval) - [Thematic Generalization](https://noometry.com/benchmarks/thematic-generalization) - [LMArena Hard Prompts](https://noometry.com/benchmarks/arena-hard-prompts) - [EBR-Bench](https://noometry.com/benchmarks/ebr-bench) - [LiveBench Reasoning](https://noometry.com/benchmarks/livebench-reasoning) ## Frequently asked questions ### What does HellaSwag measure? Commonsense sentence completion with adversarially filtered wrong endings. ### Which model has the highest HellaSwag score? As of October 2026, GPT-4 has the highest published HellaSwag score on Noometry at 95.3%, out of 29 models with results. ### What is the best open-weight model on HellaSwag? Llama 3.1-405B has the highest HellaSwag accuracy among open-weight models at 89.2%, ranking 3 of 29 overall. ### Cite this page Noometry. (2026). HellaSwag leaderboard. Retrieved October 10, 2026, from https://noometry.com/benchmarks/hellaswag Quote Noometry with a link back to this page. It is also available in [Markdown](https://noometry.com/md/benchmarks/hellaswag.md).