Multimodal Understanding on SEEDBench2 Plus
76.86AccuracyMIRROR (ours)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MIRROR (ours)Param Size=7B2026.02 | 76.86 | — | — | — | |
| GLM-9B-DeltaThinkerBackbone=GLM-9B2026.05 | 74.81 | — | — | — | |
| Qwen-8B-DeltaThinkerBackbone=Qwen-8B2026.05 | 73.43 | — | — | — | |
| GLM-4.1V-9B-ThinkingBackbone=GLM-4.1V-9B2026.05 | 73.03 | — | — | — | |
| BAGELSize=14B2026.04 | 71.9 | — | — | — | |
| Qwen3-VLparameters=8B2026.01 | 71.19 | — | — | — | |
| InternVL3.5parameters=8B2026.01 | 71.15 | — | — | — | |
| GPTQModel Backbone=Qwen2.5-VL-7B-Instruct, Quantization Method=GPTQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 71.15 | — | — | — | |
| Qwen2.5-VL-7BParam Size=7B2026.02 | 70.88 | — | — | — | |
| Full PrecModel Backbone=Qwen2.5-VL-7B-Instruct, Quantization Method=Full Precision, Quantization Bit-width/Group-size=None, Zero-shot=true2025.08 | 70.62 | — | — | — | |
| Qwen2.5-VLSize=7B2026.04 | 70.5 | — | — | — | |
| Innovator-VLvariant=8B-Instruct, parameters=8B2026.01 | 70.44 | — | — | — | |
| Intern-S1variant=mini, parameters=9B2026.01 | 70.44 | — | — | — | |
| Qwen2.5-VLLLM=Qwen2.5-7B2025.05 | 70.4 | — | — | — | |
| MIRROR (w/o tool)Param Size=7B2026.02 | 70.36 | — | — | — | |
| ResDecBase Model=Qwen2.5-VL2026.02 | 70.31 | — | — | — | |
| REVisual-R1Model Family=REVisual2026.05 | 70.14 | — | — | — | |
| DefenderIteration=32026.01 | 70.05 | 68.4 | 60.97 | 83.18 | |
| Qwen3-VL-8B-ThinkingBackbone=Qwen3-VL-8B2026.05 | 69.87 | — | — | — | |
| Innovator-VLvariant=8B-Thinking, parameters=8B2026.01 | 69.83 | — | — | — | |
| Base (M_def^(0)) + Clean DataData=Cleaned2026.01 | 69.78 | 67.53 | 61.09 | 83.18 | |
| DefenderIteration=12026.01 | 69.78 | 67.9 | 60.72 | 83.18 | |
| GPTAQModel Backbone=Qwen2.5-VL-7B-Instruct, Quantization Method=GPTAQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 69.74 | — | — | — | |
| InternVL2.5LLM=InternLM2.5-7B2025.05 | 69.7 | — | — | — | |
| DefenderIteration=22026.01 | 69.65 | 67.65 | 60.59 | 83.18 | |
| RegularBase Model=Qwen2.5-VL2026.02 | 69.61 | — | — | — | |
| InternVL3Param Size=8B2026.02 | 69.61 | — | — | — | |
| Vision-R1-7BBackbone=Vision-R1-7B2026.05 | 69.57 | — | — | — | |
| Base (M_def^(0))2026.01 | 69.52 | 67.41 | 60.1 | 83.64 | |
| VLMQModel Backbone=Qwen2.5-VL-7B-Instruct, Quantization Method=VLMQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 69.43 | — | — | — | |
| MiMo-VLtraining=7B-RL, parameters=7B2026.01 | 69.39 | — | — | — | |
| MiMo-VLtraining=7B-SFT, parameters=7B2026.01 | 69.35 | — | — | — | |
| VISTABase Model=Qwen2.5-VL2026.02 | 69.3 | — | — | — | |
| MiniCPM-Vversion=4.5, parameters=8B2026.01 | 69.26 | — | — | — | |
| Full PrecModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=Full Precision, Quantization Bit-width/Group-size=None, Zero-shot=true2025.08 | 69.26 | — | — | — | |
| OPERABase Model=Qwen2.5-VL2026.02 | 68.99 | — | — | — | |
| LLaVA-OVversion=1.5, parameters=8B2026.01 | 68.99 | — | — | — | |
| Qwen2.5-VL-3BParam Size=3B2026.02 | 68.81 | — | — | — | |
| ARES-RL-7BBackbone=ARES-RL-7B2026.05 | 68.77 | — | — | — | |
| Bee-8B-RLBackbone=Bee-8B2026.05 | 68.55 | — | — | — | |
| VCDBase Model=Qwen2.5-VL2026.02 | 67.9 | — | — | — | |
| VLMQModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=VLMQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 67.68 | — | — | — | |
| GPTAQModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=GPTAQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 67.28 | — | — | — | |
| DoLaBase Model=Qwen2.5-VL2026.02 | 67.19 | — | — | — | |
| AWQModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=AWQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 66.84 | — | — | — | |
| GPTQModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=GPTQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 66.8 | — | — | — | |
| MBQModel Backbone=Qwen2-VL-7B-Instruct, Quantization Method=MBQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 65.74 | — | — | — | |
| InternVL3Param Size=2B2026.02 | 64.95 | — | — | — | |
| Full PrecModel Backbone=LLaVA-OneVision-7B, Quantization Method=Full Precision, Quantization Bit-width/Group-size=None, Zero-shot=true2025.08 | 64.82 | — | — | — | |
| GPTAQModel Backbone=LLaVA-OneVision-7B, Quantization Method=GPTAQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 64.82 | — | — | — | |
| VLMQModel Backbone=LLaVA-OneVision-7B, Quantization Method=VLMQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 63.99 | — | — | — | |
| GPTQModel Backbone=LLaVA-OneVision-7B, Quantization Method=GPTQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 63.46 | — | — | — | |
| ICDBase Model=Qwen2.5-VL2026.02 | 62.32 | — | — | — | |
| LLaVA-OV2026.04 | 62.2 | — | — | — | |
| Full PrecModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=Full Precision, Quantization Bit-width/Group-size=None, Zero-shot=true2025.08 | 61.79 | — | — | — | |
| OneCatSize=9B2026.04 | 61.6 | — | — | — | |
| Show-o2Size=7B2026.04 | 61.3 | — | — | — | |
| Tuna-2Size=7B2026.04 | 61.1 | — | — | — | |
| Tuna-RSize=7B2026.04 | 58.4 | — | — | — | |
| GPTQModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=GPTQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 58.15 | — | — | — | |
| VLMQBase Model=Qwen2.5-VL-7B-Instruct, Bit-width=2-bit, Group Size=128, Precursor Algorithm=GPTQ2025.08 | 57.27 | — | — | — | |
| VLMQModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=VLMQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 56.92 | — | — | — | |
| Ming-UniVisionSize=16B2026.04 | 56.8 | — | — | — | |
| GPTAQModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=GPTAQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 56.79 | — | — | — | |
| Janus-ProSize=7B2026.04 | 56.3 | — | — | — | |
| MBQModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=MBQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 54.98 | — | — | — | |
| VLMQBase Model=Qwen2-VL-7B-Instruct, Bit-width=2-bit, Group Size=128, Precursor Algorithm=GPTQ2025.08 | 53.8 | — | — | — | |
| AWQModel Backbone=Qwen2-VL-2B-Instruct, Quantization Method=AWQ, Quantization Bit-width/Group-size=INT3g128, Zero-shot=true2025.08 | 53.67 | — | — | — | |
| VLMQBase Model=LLaVA-OneVision-7B, Bit-width=2-bit, Group Size=1282025.08 | 53.32 | — | — | — | |
| TunaSize=7B2026.04 | 52.7 | — | — | — | |
| GPTQBase Model=Qwen2-VL-7B-Instruct, Bit-width=2-bit, Group Size=1282025.08 | 51.78 | — | — | — | |
| GPTAQBase Model=Qwen2-VL-7B-Instruct, Bit-width=2-bit, Group Size=1282025.08 | 51.25 | — | — | — | |
| GPTQBase Model=LLaVA-OneVision-7B, Bit-width=2-bit, Group Size=1282025.08 | 51.25 | — | — | — | |
| GPTAQBase Model=LLaVA-OneVision-7B, Bit-width=2-bit, Group Size=1282025.08 | 50.94 | — | — | — | |
| GPTQBase Model=Qwen2.5-VL-7B-Instruct, Bit-width=2-bit, Group Size=1282025.08 | 48.4 | — | — | — | |
| TarSize=7B2026.04 | 46.2 | — | — | — | |
| Emu3Size=8B2026.04 | 44.6 | — | — | — | |
| SemDeDup# Data=65K2025.04 | 43.96 | — | — | — | |
| D2-Pruning# Data=65K2025.04 | 43.7 | — | — | — | |
| TIES-Merging2026.05 | 43.13 | — | — | — | |
| COMPACT# Data=65K2025.04 | 43.13 | — | — | — | |
| EL2N# Data=65K2025.04 | 42.95 | — | — | — | |
| Self-Sup# Data=65K2025.04 | 42.51 | — | — | — | |
| Task Arithmetic2026.05 | 42.07 | — | — | — | |
| ICONS# Data=65K2025.04 | 42.03 | — | — | — | |
| DARE2026.05 | 41.9 | — | — | — | |
| Perplexity# Data=65K2025.04 | 41.9 | — | — | — | |
| Random# Data=65K2025.04 | 41.85 | — | — | — | |
| LLaVA-665K# Data=665K2025.04 | 41.72 | — | — | — | |
| HarmonSize=1.5B2026.04 | 41.6 | — | — | — | |
| Self-Filter# Data=65K2025.04 | 41.33 | — | — | — | |
| ResDecBase Model=LLaVA-1.52026.02 | 41.15 | — | — | — | |
| LLaVA-v1.5Model Size=7B2026.05 | 40.97 | — | — | — | |
| MemVRBase Model=LLaVA-1.52026.02 | 40.84 | — | — | — | |
| NeuroMerging2026.05 | 40.84 | — | — | — | |
| VISTABase Model=LLaVA-1.52026.02 | 40.49 | — | — | — | |
| DiM3Model Size=7B2026.05 | 40.45 | — | — | — | |
| Breadcrumbs2026.05 | 40.32 | — | — | — | |
| JanusFlowSize=1.3B2026.04 | 39.8 | — | — | — | |
| OPERABase Model=LLaVA-1.52026.02 | 39.75 | — | — | — |