Visual Question Answering on MRAG-Bench
68.96Overall AccuracyCoGR-MoE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CoGR-MoEBackbone=Qwen3-VL-A3B-30B2026.04 | 68.96 | 74.54 | 64.61 | 62.83 | |
| MoMEBackbone=Qwen3-VL-A3B-30B2026.04 | 66.78 | 70.64 | 56.39 | 56.75 | |
| Metis-HOMEBackbone=Qwen3-VL-A3B-30B2026.04 | 65.81 | 68.45 | 61.4 | 65.3 | |
| CL-MOEBackbone=Qwen3-VL-A3B-30B2026.04 | 65.31 | 73.3 | 48.69 | 48.65 | |
| MH-MoEBackbone=Qwen3-VL-A3B-30B2026.04 | 64.85 | 65.92 | 58.82 | 50.5 | |
| I^2MoEBackbone=Qwen3-VL-A3B-30B2026.04 | 64.27 | 68.71 | 53.46 | 68 | |
| CoGR-MoEBackbone=MOE-LLaVA2026.04 | 63.25 | 69.09 | 58.55 | 58.67 | |
| MoMEBackbone=MOE-LLaVA2026.04 | 62.76 | 69.3 | 55.91 | 64.06 | |
| MH-MoEBackbone=MOE-LLaVA2026.04 | 61.71 | 69.78 | 51.7 | 40.83 | |
| I^2MoEBackbone=MOE-LLaVA2026.04 | 59.35 | 62.16 | 50.87 | 60.83 | |
| Qwen3-VL-A3B-30BBackbone=Qwen3-VL-A3B-30B2026.04 | 59.32 | 64.08 | 54.29 | 61.33 | |
| CL-MOEBackbone=MOE-LLaVA2026.04 | 58.95 | 63.15 | 51.57 | 59.94 | |
| Metis-HOMEBackbone=MOE-LLaVA2026.04 | 57.25 | 63.79 | 51.1 | 55.75 | |
| MOE-LLaVABackbone=MOE-LLaVA2026.04 | 53.29 | 55.54 | 50.78 | 51.67 |