Multimodal Reasoning on MMMU (Acc.%, Acpt.%, Speedup)
85.79AccuracyDREAM-R
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DREAM-RTarget=Q32B, Draft=Q2B2026.05 | 85.79 | 56.44 | 2.19 | |
| DREAM-RTarget=Q235B, Draft=M7B-RL2026.05 | 83.1 | 25 | 1.75 | |
| DREAM-R-NSTarget=Q235B, Draft=M7B-RL2026.05 | 81 | 23.69 | 1.65 | |
| LRTarget=Q235B, Draft=M7B-RL2026.05 | 79.8 | 23.03 | 1.45 | |
| DREAM-RTarget=Q235B, Draft=Q2B2026.05 | 78.33 | 39.21 | 1.6 | |
| DREAM-R-NSTarget=Q32B, Draft=Q2B2026.05 | 77.85 | 42.46 | 2.03 | |
| DREAM-RTarget=Q32B, Draft=Q4B2026.05 | 77.37 | 42.37 | 1.91 | |
| Standard SDTarget=Q32B, Draft=Q4B2026.05 | 77.18 | — | 1.1 | |
| Standard SDTarget=Q32B, Draft=Q2B2026.05 | 76.15 | — | 1.15 | |
| LRTarget=Q32B, Draft=Q2B2026.05 | 75.57 | 41.63 | 1.72 | |
| Standard SDTarget=Q32B, Draft=M7B-RL2026.05 | 74 | — | 1.04 | |
| DREAM-RTarget=Q32B, Draft=M7B-RL2026.05 | 73.5 | 42.1 | 1.61 | |
| DREAM-RTarget=Q235B, Draft=Q4B2026.05 | 72.53 | 57.82 | 1.58 | |
| DREAM-RTarget=Q235B, Draft=R4B2026.05 | 71.95 | 54.1 | 1.57 | |
| DREAM-R-NSTarget=Q235B, Draft=Q4B2026.05 | 71.69 | 53.29 | 1.55 | |
| Standard SDTarget=Q235B, Draft=Q2B2026.05 | 71.59 | — | 1.07 | |
| DREAM-R-NSTarget=Q235B, Draft=Q2B2026.05 | 71.59 | 37.2 | 1.52 | |
| Standard SDTarget=Q235B, Draft=M7B-RL2026.05 | 71.5 | — | 1.07 | |
| DREAM-RTarget=Q32B, Draft=R4B2026.05 | 71.4 | 53.6 | 1.98 | |
| LRTarget=Q235B, Draft=Q2B2026.05 | 71.11 | 36.66 | 1.36 | |
| DREAM-R-NSTarget=Q32B, Draft=M7B-RL2026.05 | 71.1 | 39.8 | 1.49 | |
| DREAM-R-NSTarget=Q235B, Draft=R4B2026.05 | 70.25 | 50.7 | 1.54 | |
| Standard SDTarget=Q235B, Draft=Q4B2026.05 | 69.83 | — | 1.11 | |
| LRTarget=Q32B, Draft=M7B-RL2026.05 | 69.2 | 37.3 | 1.36 | |
| DREAM-R-NSTarget=Q32B, Draft=R4B2026.05 | 69.05 | 49.2 | 1.88 | |
| LRTarget=Q235B, Draft=Q4B2026.05 | 68.2 | 53.7 | 1.48 | |
| Standard SDTarget=Q32B, Draft=R4B2026.05 | 68.15 | — | 1.08 | |
| Standard SDTarget=Q235B, Draft=R4B2026.05 | 68.05 | — | 1.1 | |
| LRTarget=Q235B, Draft=R4B2026.05 | 67.3 | 51.9 | 1.46 | |
| LRTarget=Q32B, Draft=R4B2026.05 | 67.1 | 46.9 | 1.8 | |
| DREAM-R-NSTarget=Q32B, Draft=Q4B2026.05 | 66 | 34.23 | 1.84 | |
| SpecReasonTarget=Q235B, Draft=R4B2026.05 | 60.1 | 49.8 | 1.21 | |
| SpecReasonTarget=Q235B, Draft=Q4B2026.05 | 59.72 | 51.86 | 1.22 | |
| LRTarget=Q32B, Draft=Q4B2026.05 | 58.96 | 34.35 | 1.66 | |
| SpecReasonTarget=Q235B, Draft=Q2B2026.05 | 56.71 | 35.65 | 1.07 | |
| SpecReasonTarget=Q32B, Draft=R4B2026.05 | 54.3 | 40.5 | 1.22 | |
| SpecReasonTarget=Q32B, Draft=Q4B2026.05 | 53.73 | 32.86 | 1.24 | |
| SpecReasonTarget=Q32B, Draft=M7B-RL2026.05 | 50.1 | 27.9 | 0.84 | |
| SpecReasonTarget=Q235B, Draft=M7B-RL2026.05 | 48.6 | 23.77 | 0.63 | |
| SpecReasonTarget=Q32B, Draft=Q2B2026.05 | 46.88 | 16.04 | 1.21 |