Multimodal Understanding on SEED (Metrics: SEED, Average)
81.8SEED ScoreBAGEL (Ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BAGEL (Ours)Baseline=BAGEL, Params=7B active2026.06 | 81.8 | — | |
| BLIP3o-8B (Ours)Baseline=BLIP3o-8B, Params=8B2026.06 | 79.4 | — | |
| BAGELBaseline=BAGEL, Params=7B active2026.06 | 79.3 | — | |
| VARGPT-v1.1 (Ours)Baseline=VARGPT-v1.1, Params=7B+2B2026.06 | 79.2 | — | |
| FP16Model=LLaVA-onevision-72B, Bitwidth=FP162024.12 | 78.1 | 74.3 | |
| MBQModel=Qwen2-VL-72B, Bitwidth=W3A162024.12 | 77.7 | 77.6 | |
| MBQModel=LLaVA-onevision-72B, Bitwidth=W3A162024.12 | 77.6 | 73.6 | |
| FP16Model=Qwen2-VL-72B, Bitwidth=FP162024.12 | 77.6 | 78.1 | |
| AWQModel=Qwen2-VL-72B, Bitwidth=W3A162024.12 | 77.6 | 77.5 | |
| RTNModel=Qwen2-VL-72B, Bitwidth=W3A162024.12 | 77.5 | 75 | |
| BLIP3o-8BBaseline=BLIP3o-8B, Params=8B2026.06 | 77.5 | — | |
| RTNModel=LLaVA-onevision-72B, Bitwidth=W3A162024.12 | 77.4 | 72.1 | |
| GPTQModel=Qwen2-VL-72B, Bitwidth=W3A162024.12 | 77.2 | 76.6 | |
| FP16Model=InternVL2-26B, Bitwidth=FP162024.12 | 76.8 | 74.6 | |
| RTNModel=LLaVA-onevision-72B, Bitwidth=W4A82024.12 | 76.6 | 70.7 | |
| SQModel=LLaVA-onevision-72B, Bitwidth=W4A82024.12 | 76.6 | 70.3 | |
| RTNModel=Qwen2-VL-72B, Bitwidth=W4A82024.12 | 76.6 | 73.2 | |
| MBQModel=LLaVA-onevision-72B, Bitwidth=W4A82024.12 | 76.5 | 70.8 | |
| SQModel=Qwen2-VL-72B, Bitwidth=W4A82024.12 | 76.4 | 72.2 | |
| MBQModel=InternVL2-26B, Bitwidth=W3A162024.12 | 76.3 | 73.8 | |
| MBQModel=Qwen2-VL-72B, Bitwidth=W4A82024.12 | 76.3 | 75.8 | |
| AWQModel=InternVL2-26B, Bitwidth=W3A162024.12 | 76.2 | 73.5 | |
| VARGPT-v1.1Baseline=VARGPT-v1.1, Params=7B+2B2026.06 | 76.1 | — | |
| GPTQModel=LLaVA-onevision-72B, Bitwidth=W3A162024.12 | 76 | 72.3 | |
| RTNModel=InternVL2-26B, Bitwidth=W4A82024.12 | 76 | 72.7 | |
| GPTQModel=InternVL2-26B, Bitwidth=W3A162024.12 | 75.8 | 72.3 | |
| RTNModel=InternVL2-26B, Bitwidth=W3A162024.12 | 75.7 | 73.3 | |
| MBQModel=InternVL2-26B, Bitwidth=W4A82024.12 | 75.7 | 72.7 | |
| Lumina-DiMOO (ReMDM)2026.02 | 74.3 | — | |
| FreeCorrectionbackbone=Lumina-DiMOO2026.02 | 74.2 | — | |
| ILLUMEBaseline=Vicuna-7B, Params=7B2026.06 | 72.9 | — | |
| TokenFlowBaseline=TokenFlow2026.06 | 72.6 | — | |
| Janus-Pro-7BBaseline=Janus-Pro-7B, Params=7B2026.06 | 72.1 | — | |
| SUDERBaseline=Janus-Pro-7B, Params=7B2026.06 | 71.9 | — | |
| MetaMorphBaseline=MetaMorph2026.06 | 71.8 | — | |
| Lumina-DiMOOgeneration step=64, generated sequence length=2562026.02 | 71.4 | — | |
| AWQModel=LLaVA-onevision-72B, Bitwidth=W3A162024.12 | 71.2 | 55.1 | |
| SQModel=InternVL2-26B, Bitwidth=W4A82024.12 | 70.6 | 65.5 | |
| Emu3Baseline=Emu32026.06 | 68.2 | — | |
| SEED-XBaseline=SEED-X2026.06 | 66.5 | — | |
| JanusBaseline=Janus2026.06 | 63.7 | — | |
| VILA-UBaseline=VILA-U2026.06 | 59 | — | |
| LLaVA-1.5-7BTokens=5762026.05 | 57.9 | 100 | |
| ETCTokens=42026.05 | 56.66 | 95.33 | |
| VoCoTokens=42026.05 | 56 | — | |
| ETCTokens=22026.05 | 55.76 | 94.55 | |
| VoCoTokens=22026.05 | 55 | — | |
| ETCTokens=12026.05 | 54.18 | 92.22 | |
| VoCoTokens=12026.05 | 53.7 | — | |
| LLaVA-1.5-7BTokens=02026.05 | 29.2 | — | |
| ChameleonBaseline=Chameleon2026.06 | 27.2 | — |