MathVista — Full MathVista test set; distinct subset axis from MathVista (mini) (testmini split).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi K2.5 | Moonshot | 90.1 | 2 | 2026-08-23 |
| 2 | Gemini 3 Pro | 89.8 | 1 | 2026-06-15 | |
| 3 | Qwen3.5 27B | Alibaba | 87.8 | 3 | 2026-08-24 |
| 4 | Qwen3.6 27B | Alibaba | 87.4 | 2 | 2026-08-24 |
| 5 | Qwen3.5 122B A10B | Alibaba | 87.4 | 1 | 2026-08-23 |
| 6 | o3 | OpenAI | 86.8 | 1 | 2026-08-23 |
| 7 | Qwen3.6 35B A3B | Alibaba | 86.4 | 2 | 2026-08-24 |
| 8 | Qwen3.5 35B A3B | Alibaba | 86.2 | 2 | 2026-08-24 |
| 9 | Qwen3 VL 32B Reasoning | Alibaba | 85.9 | 1 | 2026-08-23 |
| 10 | Qwen3 VL 235B A22B Reasoning | Alibaba | 85.8 | 3 | 2026-08-23 |
| 11 | Step3 VL 10B | StepFun | 85.5 | 3 | 2026-08-23 |
| 12 | Qwen3 VL 235B A22B Instruct | Alibaba | 84.9 | 1 | 2026-08-23 |
| 13 | o4-mini | OpenAI | 84.3 | 1 | 2026-08-23 |
| 14 | Gemini 2.5 Pro | 83.9 | 2 | 2026-06-05 | |
| 15 | Qwen3 VL 32B Instruct | Alibaba | 83.8 | 1 | 2026-08-23 |
| 16 | GLM-4.6V (106B-A12B) | Z.ai | 83.5 | 2 | 2026-06-05 |
| 17 | GLM-4.6V-Flash (9B) | Z.ai | 82.8 | 2 | 2026-06-05 |
| 18 | GPT-5.2 | OpenAI | 82.8 | 1 | 2026-06-15 |
| 19 | Qwen3 VL 30B A3B Reasoning | Alibaba | 81.9 | 1 | 2026-08-23 |
| 20 | Qwen3 VL Thinking (8B) | Alibaba | 81.4 | 3 | 2026-08-23 |
| 21 | Command A+ | Cohere | 80.6 | 1 | 2026-08-23 |
| 22 | Claude Opus 4.5 | Anthropic | 80.2 | 1 | 2026-06-15 |
| 23 | Qwen3 VL 30B A3B Instruct | Alibaba | 80.1 | 1 | 2026-08-23 |
| 24 | MiMo VL RL 2508 (7B) | Xiaomi | 79.9 | 2 | 2026-06-05 |
| 25 | Claude Sonnet 4.5 | Anthropic | 79.8 | 1 | 2026-08-24 |
| 26 | Qwen3 VL 4B (Reasoning) | Alibaba | 79.5 | 1 | 2026-08-23 |
| 27 | Gemma 4 26B A4B | 79.4 | 1 | 2026-08-24 | |
| 28 | Gemma 4 31B | 79.3 | 2 | 2026-08-24 | |
| 29 | Qwen3 VL 8B Instruct | Alibaba | 77.2 | 1 | 2026-08-23 |
| 30 | InternVL-3.5 (8B) | OpenGVLab | 76.8 | 2 | 2026-06-05 |
| 31 | Qwen2.5 VL 72B | Alibaba | 74.8 | 1 | 2026-08-23 |
| 32 | Qwen2.5 VL 32B | Alibaba | 74.7 | 1 | 2026-08-23 |
| 33 | Qwen3 VL 4B Instruct | Alibaba | 73.7 | 1 | 2026-08-23 |
| 34 | Llama 4 Maverick | Meta | 73.7 | 2 | 2026-08-23 |
| 35 | Gemini 2.0 Flash Exp | 73.1 | 1 | 2026-06-05 | |
| 36 | GPT-4.1 mini | OpenAI | 73.1 | 1 | 2026-08-23 |
| 37 | GPT-4.5 Preview | OpenAI | 72.3 | 1 | 2026-08-23 |
| 38 | GPT-4.1 | OpenAI | 72.2 | 1 | 2026-08-23 |
| 39 | o1 | OpenAI | 71.8 | 1 | 2026-08-23 |
| 40 | Llama 4 Scout | Meta | 70.7 | 3 | 2026-08-23 |
| 41 | Gemini 1.5 Pro | 70.6 | 4 | 2026-08-24 | |
| 42 | Qwen2-VL-72B-Instruct | Alibaba | 70.5 | 2 | 2026-08-24 |
| 43 | Pixtral Large | Mistral | 69.4 | 1 | 2026-08-23 |
| 44 | Grok 2 | SpaceXAI | 69 | 1 | 2026-08-23 |
| 45 | Kimi VL A3B Instruct | Moonshot | 68.7 | 1 | 2026-06-12 |
| 46 | MiniMax VL 01 | MiniMax | 68.6 | 1 | 2026-06-05 |
| 47 | LFM2.5-VL-3B | Liquid AI | 68.5 | 3 | 2026-08-23 |
| 48 | LFM2-VL-3B (3.1B) | Liquid AI | 68.5 | 1 | 2026-08-12 |
| 49 | InternVL2.5-78B | OpenGVLab | 68.4 | 1 | 2026-06-05 |
| 50 | Grok 2 Mini | SpaceXAI | 68.1 | 1 | 2026-08-23 |
| 51 | Qwen2.5 Omni 7B | Alibaba | 67.9 | 1 | 2026-08-23 |
| 52 | Claude 3.5 Sonnet | Anthropic | 67.7 | 5 | 2026-08-24 |
| 53 | Gemma 3 27B | 67.6 | 1 | 2026-08-23 | |
| 54 | GPT-4o | OpenAI | 63.8 | 5 | 2026-08-24 |
| 55 | Gemma 3 12B | 62.9 | 2 | 2026-08-23 | |
| 56 | Phi 4 Multimodal Instruct | Microsoft | 62.4 | 1 | 2026-08-23 |
| 57 | LFM2-VL-3B | Liquid AI | 62.1 | 1 | 2026-08-22 |
| 58 | Gemma 4 E2B | 62.1 | 1 | 2026-08-12 | |
| 59 | Llama 3.2 Instruct 90B (Vision) | Meta | 57.3 | 1 | 2026-06-05 |
| 60 | GPT-4o mini | OpenAI | 56.7 | 3 | 2026-08-24 |
| 61 | GPT-4.1 nano | OpenAI | 56.2 | 1 | 2026-08-23 |
| 62 | Gemini 1.5 Flash 8B | 54.7 | 1 | 2026-08-23 | |
| 63 | InternVL-2-4B | OpenGVLab | 53.7 | 1 | 2026-08-24 |
| 64 | InternVL-2-8B | OpenGVLab | 51.1 | 1 | 2026-08-24 |
| 65 | Gemma 3 4B | 50 | 1 | 2026-08-23 | |
| 66 | Phi 3.5 Vision Instruct | Microsoft | 43.9 | 2 | 2026-08-24 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.