General Image Understanding on MMStar
72.13AccuracyDPE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DPEBackbone=Qwen3-VL-8B-Instruct, Iteration=32026.02 | 72.13 | 64.39 | |
| Vanilla MoEBackbone=QWEN3-VL-30B-A3B-INSTRUCT2026.04 | 72.1 | — | |
| Vanilla MoEBackbone=INTERNVL3.5-30B-A3B2026.04 | 72.03 | — | |
| MACS (Ours)Backbone=QWEN3-VL-30B-A3B-INSTRUCT2026.04 | 71.93 | — | |
| MACS (Ours)Backbone=INTERNVL3.5-30B-A3B2026.04 | 71.84 | — | |
| MACS (w/o Expanded)Backbone=QWEN3-VL-30B-A3B-INSTRUCT2026.04 | 71.48 | — | |
| MACS (w/o Expanded)Backbone=INTERNVL3.5-30B-A3B2026.04 | 71.36 | — | |
| Qwen2.5-VL-72BParameters=72B2026.02 | 70.8 | 61.9 | |
| CAI-MoE (Expanded)Backbone=QWEN3-VL-30B-A3B-INSTRUCT2026.04 | 68.42 | — | |
| Qwen2.5-VLParam Size=32B2026.07 | 68.4 | — | |
| CAI-MoE (Expanded)Backbone=INTERNVL3.5-30B-A3B2026.04 | 67.91 | — | |
| Claude4-Sonnet2026.02 | 67.4 | 64.1 | |
| CAI-MoE (Token Drop)Backbone=QWEN3-VL-30B-A3B-INSTRUCT2026.04 | 66.58 | — | |
| GPT-4oParam Size=-2026.07 | 65.7 | — | |
| CAI-MoE (Token Drop)Backbone=INTERNVL3.5-30B-A3B2026.04 | 65.47 | — | |
| GPT-4o2026.02 | 64.7 | 56.1 | |
| Hybrid-Frame StrategyBackbone=Qwen2.5-VL-7B, Frames=8F2026.03 | 62.33 | — | |
| Hybrid-Frame StrategyBackbone=Qwen2.5-VL-7B, Frames=16F2026.03 | 62.33 | — | |
| Hybrid-Frame StrategyBackbone=Qwen2.5-VL-7B, Frames=32F2026.03 | 62.33 | — | |
| Hybrid-Frame StrategyBackbone=Qwen2.5-VL-7B, Frames=64F2026.03 | 62.33 | — | |
| SaGe-7BParam Size=7B2026.07 | 62.3 | — | |
| Video-SFTBackbone=Qwen2.5-VL-7B, Frames=64F2026.03 | 62.27 | — | |
| Video-SFTBackbone=Qwen2.5-VL-7B, Frames=32F2026.03 | 62.2 | — | |
| Video-SFTBackbone=Qwen2.5-VL-7B, Frames=16F2026.03 | 62 | — | |
| Vision-R1Param Size=7B2026.07 | 61.8 | — | |
| LLaVA-OVParam Size=7B2026.07 | 61.7 | — | |
| Video-SFTBackbone=Qwen2.5-VL-7B, Frames=8F2026.03 | 61.67 | — | |
| GPT5-Mini2026.02 | 61.3 | 53.8 | |
| Vanilla MoEBackbone=KIMI-VL-A3B-INSTRUCT2026.04 | 61.25 | — | |
| MACS (Ours)Backbone=KIMI-VL-A3B-INSTRUCT2026.04 | 61.16 | — | |
| LVRParam Size=7B2026.07 | 61.1 | — | |
| MACS (w/o Expanded)Backbone=KIMI-VL-A3B-INSTRUCT2026.04 | 60.89 | — | |
| Qwen2.5-VLParam Size=7B2026.07 | 60.3 | — | |
| DPA-Qwen3-32BTarget LLM=Qwen3-32B2026.05 | 59.1 | — | |
| CAI-MoE (Expanded)Backbone=KIMI-VL-A3B-INSTRUCT2026.04 | 58.17 | — | |
| LFM2-VL-3B# Total Params=3.0B, Evaluation Toolkit=VLMEvalKit2025.11 | 57.73 | — | |
| InternVL3.5-2B# Total Params=2.3B, Evaluation Toolkit=VLMEvalKit2025.11 | 57.67 | — | |
| CAI-MoE (Token Drop)Backbone=KIMI-VL-A3B-INSTRUCT2026.04 | 56.83 | — | |
| Qwen2.5-VL-3B# Total Params=3.75B, Evaluation Toolkit=VLMEvalKit2025.11 | 56.13 | — | |
| LLaVA-NeXT-Qwen3-32BTarget LLM=Qwen3-32B2026.05 | 56.1 | — | |
| SaGe-3BParam Size=3B2026.07 | 54.4 | — | |
| Qwen2.5-VLParam Size=3B2026.07 | 52.2 | — | |
| Cambrian-1-8BTarget LLM=Public Models2026.05 | 50.9 | — | |
| LLaVA-NeXT-Qwen3-4BTarget LLM=Qwen3-4B2026.05 | 50.7 | — | |
| DPA-Qwen3-4BTarget LLM=Qwen3-4B2026.05 | 50.7 | — | |
| InternVL3.5-1B# Total Params=1.1B2025.11 | 50.27 | — | |
| Kimi-VL-A3B-Instructk=6, Status=Default2025.11 | 49.89 | — | |
| LFM2-VL-1.6B# Total Params=1.6B2025.11 | 49.87 | — | |
| Idefics2-8BTarget LLM=Public Models2026.05 | 49.4 | — | |
| Qwen3-VL-2B# Total Params=2.1B, Evaluation Toolkit=VLMEvalKit2025.11 | 49 | — | |
| Qwen2-VL-2BTarget LLM=Public Models2026.05 | 48 | — | |
| MoDESSkip x% Experts=83%, rho=0.82025.11 | 46.68 | — | |
| SmolVLM2-2.2B# Total Params=2.2B, Evaluation Toolkit=VLMEvalKit2025.11 | 46 | — | |
| LLaVA-NeXT-LLaMA-3.2-3BTarget LLM=LLaMA-3.2-3B2026.05 | 41 | — | |
| LFM2-VL-450M# Total Params=0.45B2025.11 | 40.87 | — | |
| DPA-LLaMA-3.2-3BTarget LLM=LLaMA-3.2-3B2026.05 | 40.7 | — | |
| SmolVLM2-500M# Total Params=0.5B2025.11 | 37.73 | — | |
| LLaVA-NeXT-7BTarget LLM=Public Models2026.05 | 37.7 | — | |
| UnicornText-only=true, Pretrain Scale=1.2M2025.03 | 35.13 | — | |
| BunnyText-only=false, Pretrain Scale=1.2M2025.03 | 34.73 | — | |
| Unicorn*Text-only=true, Pretrain Scale=558K2025.03 | 34.2 | — | |
| LLaVA-1.5Text-only=false, Pretrain Scale=558K2025.03 | 33.27 | — | |
| Hybrid-Frame StrategyBackbone=LLaVA-1.5-7B, Note=Evaluated with a fixed 8-frame inference2026.03 | 32.47 | — | |
| LLaVA-1.5-7BTarget LLM=Public Models2026.05 | 32.3 | — | |
| Video-SFTBackbone=LLaVA-1.5-7B, Frames=32F2026.03 | 29.4 | — | |
| Video-SFTBackbone=LLaVA-1.5-7B, Frames=8F2026.03 | 28.87 | — | |
| Video-SFTBackbone=LLaVA-1.5-7B, Frames=16F2026.03 | 28.47 | — |