BIG-Bench Hard — Anchor: bbh=BIG-Bench Hard, NOT BigLaw/BioMystery (acronym false-positives), NOT harder BBEH variant
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude 3.5 Sonnet | Anthropic | 93.1 | 1 | 2026-08-23 |
| 2 | Hunyuan Large Inst. | Tencent | 89.5 | 1 | 2026-05-31 |
| 3 | Gemini 1.5 Pro | 89.2 | 1 | 2026-08-23 | |
| 4 | Qwen3 235B A22B | Alibaba | 88.9 | 1 | 2026-08-23 |
| 5 | Kimi K2 Base | Moonshot | 88.7 | 7 | 2026-06-13 |
| 6 | DeepSeek-V3.2-Exp-Base | DeepSeek | 88.7 | 4 | 2026-06-13 |
| 7 | MiMo V2 Flash Base | Xiaomi | 88.5 | 4 | 2026-06-13 |
| 8 | MiMo V2.5 Pro Base | Xiaomi | 88.4 | 3 | 2026-06-05 |
| 9 | MiMo V2.5 Pro | Xiaomi | 88.4 | 1 | 2026-08-23 |
| 10 | DeepSeek-R1-Distill-Qwen-32B | DeepSeek | 88.3 | 1 | 2026-06-05 |
| 11 | DeepSeek-V3.1-Base | DeepSeek | 88.2 | 4 | 2026-06-13 |
| 12 | Step3.5 Flash Base | StepFun | 88.2 | 2 | 2026-06-05 |
| 13 | Gemma 3 27B | 87.6 | 1 | 2026-08-23 | |
| 14 | DeepSeek-V4-Pro-Base | DeepSeek | 87.5 | 3 | 2026-06-05 |
| 15 | DeepSeek-V3 | DeepSeek | 87.5 | 1 | 2026-05-01 |
| 16 | MiMo V2.5 Base | Xiaomi | 87.2 | 3 | 2026-06-05 |
| 17 | Nova Pro (Non-Reasoning) | Amazon | 86.9 | 1 | 2026-08-23 |
| 18 | DeepSeek-V4-Flash-Base | DeepSeek | 86.9 | 3 | 2026-06-05 |
| 19 | Claude 3 Opus | Anthropic | 86.8 | 1 | 2026-08-23 |
| 20 | Hunyuan Large | Tencent | 86.3 | 1 | 2026-05-31 |
| 21 | GLM-4.5-Base | Z.ai | 86.2 | 2 | 2026-06-05 |
| 22 | Llama 3.1 Instruct 405B | Meta | 85.9 | 2 | 2026-05-31 |
| 23 | SynLogic 32B | MiniMax | 85.8 | 1 | 2026-06-05 |
| 24 | Gemma 3 12B | 85.7 | 1 | 2026-08-23 | |
| 25 | Qwen2.5 Instruct 32B | Alibaba | 84.5 | 1 | 2026-06-05 |
| 26 | DeepSeek-V2.5 | DeepSeek | 84.3 | 2 | 2026-08-23 |
| 27 | Granite 4.1 30B | IBM | 83.7 | 1 | 2026-06-15 |
| 28 | Claude 3 Sonnet | Anthropic | 82.9 | 1 | 2026-08-23 |
| 29 | Qwen2 Instruct (72B) | Alibaba | 82.4 | 1 | 2026-08-23 |
| 30 | Nova Lite (Non-Reasoning) | Amazon | 82.4 | 1 | 2026-08-23 |
| 31 | Llama 3.1 70B Instruct | Meta | 81.6 | 1 | 2026-05-31 |
| 32 | Granite 4.1 30B Base | IBM | 80.7 | 2 | 2026-06-15 |
| 33 | Granite 4.1 8B | IBM | 80.5 | 1 | 2026-06-15 |
| 34 | Qwen2.5 Instruct 72B | Alibaba | 79.8 | 1 | 2026-05-01 |
| 35 | Nova Micro (Non-Reasoning) | Amazon | 79.5 | 1 | 2026-08-23 |
| 36 | Phi 3.5 MoE Instruct | Microsoft | 79.1 | 1 | 2026-08-23 |
| 37 | DeepSeek-V2 | DeepSeek | 78.9 | 2 | 2026-05-31 |
| 38 | Granite 4.1 8B Base | IBM | 73.8 | 2 | 2026-06-15 |
| 39 | Claude 3 Haiku | Anthropic | 73.7 | 1 | 2026-08-23 |
| 40 | Gemma 3 4B | 72.2 | 1 | 2026-08-23 | |
| 41 | Qwen2.5 7B Base | Alibaba | 70.4 | 1 | 2026-05-03 |
| 42 | Phi 4 Mini Instruct | Microsoft | 70.4 | 1 | 2026-08-23 |
| 43 | Granite 3.3 8B Instruct | IBM | 69.1 | 1 | 2026-08-23 |
| 44 | Phi 3.5 Mini Instruct | Microsoft | 69 | 1 | 2026-08-23 |
| 45 | SynLogic 7B | MiniMax | 66.5 | 1 | 2026-06-05 |
| 46 | Moonlight | Moonshot | 65.2 | 2 | 2026-06-12 |
| 47 | Granite 4.1 3B Base | IBM | 63.8 | 1 | 2026-06-15 |
| 48 | Qwen2.5 Omni 3B | Alibaba | 61.3 | 1 | 2026-08-09 |
| 49 | Qwen2.5 3B Base | Alibaba | 59.5 | 1 | 2026-05-03 |
| 50 | Qwen2.5 3B | Alibaba | 56.3 | 2 | 2026-06-05 |
| 51 | Qwen3 4B Base Dense | Alibaba | 54.5 | 1 | 2026-05-03 |
| 52 | Llama 3.2 3B Instruct | Meta | 46.8 | 2 | 2026-06-05 |
| 53 | Gemma 3 1B | 39.1 | 1 | 2026-08-23 | |
| 54 | ERNIE 4.5 | Baidu | 30.4 | 1 | 2026-08-23 |
| 55 | Gemma 3 1B Base Dense | 26.3 | 1 | 2026-05-03 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.