LogicVista
Multimodal · 2024-07-06
LogicVista evaluates logical reasoning in visual contexts with 448 multiple-choice questions across five reasoning skills (inductive, deductive, numerical, spatial and mechanical) and nine capabilities. Scores report overall accuracy.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemma 4 12B IT | 76.3 |
| GPT-5.6 Luna | 74.7 |
| Qwen3 VL 235B A22B Thinking | 72.2 |
| Gemini 2.5 Pro | 72.0 |
| GPT-5 | 71.8 |
| GPT-5 Mini | 71.4 |
| Qwen3 VL 32B Thinking | 70.9 |
| Qwen3.5-4B | 69.8 |
| Opus 4.1 | 67.3 |
| Gemini 2.5 Flash | 67.3 |
| Qwen3-VL-235B-A22B-Instruct | 65.8 |
| Qwen3 VL 30B A3B Thinking | 65.8 |
| Qwen3 VL 8B Thinking | 65.1 |
| Qwen3 VL 32B Instruct | 62.2 |
| Qwen3 VL 4B Thinking | 61.1 |
| LFM3-VL-2B | 60.4 |
| GPT-5 Nano | 59.7 |
| Gemini 3.5 Flash-Lite | 57.0 |
| Qwen3-VL-8B-Instruct | 55.3 |
| Qwen3.5-2B | 54.4 |
| Qwen3 VL 30B A3B Instruct | 53.5 |
| Qwen3 VL 4B Instruct | 53.2 |
| Qwen3 VL 2B Thinking | 50.0 |
| Gemma 4 E2B IT | 48.3 |
| LFM2.5-VL-3B | 37.4 |
| Qwen3 VL 2B Instruct | 35.8 |