Arena-Hard-Auto — Generic/unspecified variant; suspicious 0.3-92.4 range suggests mixed units, but identity clear
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen3 235B A22B | Alibaba | 95.6 | 1 | 2026-08-23 |
| 2 | Qwen3 32B | Alibaba | 93.8 | 1 | 2026-08-23 |
| 3 | GPT-4o | OpenAI | 92.4 | 1 | 2026-06-06 |
| 4 | DeepSeek-V3 | DeepSeek | 91.4 | 1 | 2026-06-06 |
| 5 | Qwen3 30B A3B | Alibaba | 91 | 1 | 2026-08-23 |
| 6 | MiniMax Text 01 | MiniMax | 89.1 | 1 | 2026-06-06 |
| 7 | Claude 3.5 Sonnet | Anthropic | 87.6 | 1 | 2026-06-06 |
| 8 | Gemini 1.5 Pro | 85.3 | 1 | 2026-06-06 | |
| 9 | Hunyuan Large Inst. | Tencent | 81.8 | 1 | 2026-05-31 |
| 10 | Qwen2.5 Instruct 72B | Alibaba | 81.2 | 2 | 2026-08-23 |
| 11 | Phi 4 Reasoning Plus | Microsoft | 79 | 1 | 2026-08-23 |
| 12 | DeepSeek-V2.5 | DeepSeek | 76.2 | 2 | 2026-08-23 |
| 13 | Phi 4 | Microsoft | 75.4 | 1 | 2026-08-23 |
| 14 | Gemini 2.0 Flash Exp | 72.7 | 1 | 2026-06-06 | |
| 15 | Granite 4.1 30B | IBM | 71 | 1 | 2026-06-15 |
| 16 | Llama 3.1 Instruct 405B | Meta | 69.3 | 2 | 2026-06-06 |
| 17 | Granite 4.1 8B | IBM | 69 | 1 | 2026-06-15 |
| 18 | Jamba 1.5 Large | AI21 | 65.4 | 1 | 2026-08-23 |
| 19 | Mistral Small 4 | Mistral | 58.3 | 1 | 2026-08-23 |
| 20 | Granite 3.3 8B Instruct | IBM | 57.6 | 2 | 2026-08-23 |
| 21 | Llama 3.1 70B Instruct | Meta | 55.7 | 1 | 2026-05-31 |
| 22 | Granite 3.2 8B Instruct | IBM | 55.3 | 1 | 2026-06-11 |
| 23 | Ministral 3 14B | Mistral | 55.1 | 4 | 2026-06-12 |
| 24 | Mistral Large 3 | Mistral | 55.1 | 1 | 2026-08-23 |
| 25 | Qwen3 VL 8B Instruct | Alibaba | 52.8 | 4 | 2026-06-05 |
| 26 | Ministral 3 8B | Mistral | 50.9 | 4 | 2026-06-12 |
| 27 | Jamba 1.5 Mini | AI21 | 46.1 | 1 | 2026-08-23 |
| 28 | Qwen3 VL 4B Instruct | Alibaba | 43.8 | 4 | 2026-06-05 |
| 29 | Gemma 3 12B | 43.6 | 4 | 2026-06-05 | |
| 30 | Qwen3 14B | Alibaba | 42.7 | 4 | 2026-06-05 |
| 31 | Phi 3.5 MoE Instruct | Microsoft | 37.9 | 1 | 2026-08-23 |
| 32 | Granite 3.1 8B Instruct | IBM | 37.6 | 1 | 2026-06-11 |
| 33 | Phi 3.5 Mini Instruct | Microsoft | 37 | 1 | 2026-08-23 |
| 34 | Llama 3.1 8B Instruct | Meta | 36.4 | 1 | 2026-06-11 |
| 35 | Phi 4 Mini Instruct | Microsoft | 32.8 | 1 | 2026-08-23 |
| 36 | Gemma 3 4B | 31.8 | 4 | 2026-06-05 | |
| 37 | Ministral 3 3B | Mistral | 30.5 | 4 | 2026-06-12 |
| 38 | Granite 3.3 2B Instruct | IBM | 28.9 | 1 | 2026-06-11 |
| 39 | Granite 3.2 2B Instruct | IBM | 24.9 | 1 | 2026-06-11 |
| 40 | Granite 3.1 2B Instruct | IBM | 23.3 | 1 | 2026-06-11 |
| 41 | DeepSeek-R1-Distill-Llama-8B | DeepSeek | 17.2 | 1 | 2026-06-11 |
| 42 | DeepSeek-R1-Distill-Qwen-7B | DeepSeek | 10.4 | 1 | 2026-06-11 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.