HellaSwag (commonsense NLI benchmark) — Standard accuracy-based HellaSwag; distinct source cluster from HellaSwag (EM).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude 3 Opus | Anthropic | 95.4 | 1 | 2026-08-23 |
| 2 | GPT4 | OpenAI | 95.3 | 1 | 2026-08-23 |
| 3 | Kimi K2 Base | Moonshot | 94.6 | 8 | 2026-06-13 |
| 4 | Gemini 1.5 Pro | 93.3 | 1 | 2026-08-23 | |
| 5 | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 | NVIDIA | 90.5 | 1 | 2026-08-24 |
| 6 | Nemotron 3 Ultra 550B A55B Base | NVIDIA | 90.5 | 1 | 2026-06-05 |
| 7 | DeepSeek-V2.5 | DeepSeek | 90.3 | 1 | 2026-05-31 |
| 8 | Step3.5 Flash Base | StepFun | 90.2 | 2 | 2026-06-05 |
| 9 | GLM-4.5-Base | Z.ai | 90.2 | 3 | 2026-06-05 |
| 10 | Qwen3.6 27B | Alibaba | 90 | 1 | 2026-05-01 |
| 11 | MiMo V2.5 Pro Base | Xiaomi | 89.8 | 3 | 2026-06-05 |
| 12 | MiMo V2.5 Pro | Xiaomi | 89.8 | 1 | 2026-08-23 |
| 13 | DeepSeek-V3.2-Exp-Base | DeepSeek | 89.4 | 5 | 2026-06-13 |
| 14 | Llama 3.1 Instruct 405B | Meta | 89.2 | 1 | 2026-05-01 |
| 15 | DeepSeek-V3.1-Base | DeepSeek | 89.2 | 4 | 2026-06-13 |
| 16 | Claude 3 Sonnet | Anthropic | 89 | 1 | 2026-08-23 |
| 17 | Nemotron 3 Super 120B A12B | NVIDIA | 89 | 1 | 2026-08-11 |
| 18 | DeepSeek-V3 | DeepSeek | 88.9 | 1 | 2026-05-01 |
| 19 | Mistral Large 3 675B Base 2512 | Mistral | 88.9 | 1 | 2026-06-05 |
| 20 | Command R+ (Apr '24) | Cohere | 88.6 | 1 | 2026-08-23 |
| 21 | MiMo V2.5 Base | Xiaomi | 88.6 | 3 | 2026-06-05 |
| 22 | MiMo V2 Flash Base | Xiaomi | 88.5 | 4 | 2026-06-13 |
| 23 | Hunyuan Large Inst. | Tencent | 88.5 | 1 | 2026-05-31 |
| 24 | DeepSeek-V4-Pro-Base | DeepSeek | 88 | 3 | 2026-06-05 |
| 25 | DeepSeek-V2 | DeepSeek | 87.8 | 2 | 2026-05-31 |
| 26 | Qwen2 Instruct (72B) | Alibaba | 87.6 | 1 | 2026-08-23 |
| 27 | Hunyuan Large | Tencent | 86.8 | 1 | 2026-05-31 |
| 28 | Gemma 2 27B IT | 86.4 | 1 | 2026-08-23 | |
| 29 | Claude 3 Haiku | Anthropic | 85.9 | 1 | 2026-08-23 |
| 30 | DeepSeek-V4-Flash-Base | DeepSeek | 85.7 | 3 | 2026-06-05 |
| 31 | Qwen3.5 35B A3B | Alibaba | 85.6 | 1 | 2026-08-11 |
| 32 | Llama 3.1 Nemotron Instruct 70B | NVIDIA | 85.6 | 1 | 2026-08-23 |
| 33 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16 | NVIDIA | 85.5 | 1 | 2026-08-24 |
| 34 | Nemotron 3.5 Lightning | NVIDIA | 85.5 | 1 | 2026-08-11 |
| 35 | NVIDIA Nemotron 3 Nano 30B A3B | NVIDIA | 85.5 | 1 | 2026-08-11 |
| 36 | Gemma 4 26B A4B | 85.3 | 1 | 2026-08-11 | |
| 37 | Qwen2.5 Instruct 72B | Alibaba | 84.8 | 1 | 2026-05-01 |
| 38 | Phi 3.5 MoE Instruct | Microsoft | 83.8 | 1 | 2026-08-23 |
| 39 | Qwen 2.5 Coder 32B Instruct | Alibaba | 83 | 1 | 2026-08-23 |
| 40 | Gemma 2 Instruct (9B) | 81.9 | 1 | 2026-08-23 | |
| 41 | Llama 3.2 3B Instruct | Meta | 69.8 | 1 | 2026-08-23 |
| 42 | Phi 3.5 Mini Instruct | Microsoft | 69.4 | 1 | 2026-08-23 |
| 43 | Phi 4 Mini Instruct | Microsoft | 69.1 | 1 | 2026-08-23 |
| 44 | ERNIE 4.5 | Baidu | 33 | 1 | 2026-08-23 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.