Multimodal Knowledge and Math on MMMU (val)
69.1AccuracyGPT-4o
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-4o2024.10 | 69.1 | |
| Gemini-1.5 Pro2024.10 | 62.2 | |
| Ovis2.5Parameter Scale=9B2026.01 | 60.6 | |
| GPT-4o mini2024.10 | 59.4 | |
| Qwen3-VLParameter Scale=8B2026.01 | 57.1 | |
| GPT-4V2024.10 | 56.4 | |
| Gemini-1.5 Flash2024.10 | 56.1 | |
| ARIA2024.10 | 54.9 | |
| Ostrakon-VLParameter Scale=8B2026.01 | 54.8 | |
| InternVL3.5Parameter Scale=8B2026.01 | 54.4 | |
| GLM-4.6V-FlashX2026.01 | 54.2 | |
| Pixtral-12B2024.10 | 52.5 | |
| Qwen2.5-VLParameter Scale=72B2026.01 | 51.5 | |
| Llama3.2-11B2024.10 | 50.7 | |
| LLaVA-OV-7B + HERMESKV Size=1002026.05 | 48 | |
| Cambrian-4B + Gaze attentionKV Size=288+42026.05 | 47.4 | |
| Qwen2.5-VL-3BKV Size=2562026.05 | 46.6 | |
| Cambrian-4BKV Size=5762026.05 | 46.4 | |
| Cambrian-4B + Gaze attentionKV Size=72+42026.05 | 46.4 | |
| Cambrian-4B + Gaze attentionKV Size=144+42026.05 | 46.3 | |
| LLaVA-OV-7B + HERMESKV Size=402026.05 | 46 | |
| Cambrian-4B + TokenCompre.KV Size=1442026.05 | 45.9 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=1282026.05 | 45.8 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=36+42026.05 | 45.7 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=54+42026.05 | 45.3 | |
| LLaVA-OV-7BKV Size=1962026.05 | 45.1 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=502026.05 | 44.2 | |
| LLaVA-Next-LLaMA-3-8BKV Size=7842026.05 | 39.6 | |
| Cambrian-8B-DINOv2-LKV Size=5762026.05 | 36.3 | |
| LLaVA-Next-Qwen2-7BKV Size=7842026.05 | 36.1 | |
| Cambrian-8B-CLIP-LKV Size=5762026.05 | 36.1 | |
| SPHINX-7B2026.05 | 31.6 | |
| BLIP-2-7BKV Size=5762026.05 | 26.8 |