Multimodal Reasoning on RealWorldQA (Acc, Acpt, Speedup)
81.39AccuracyDREAM-R
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DREAM-RTarget=Q32B, Draft=Q4B2026.05 | 81.39 | 52.36 | 1.8 | |
| DREAM-R-NSTarget=Q32B, Draft=Q4B2026.05 | 81.38 | 47.43 | 1.71 | |
| DREAM-RTarget=Q235B, Draft=M7B-RL2026.05 | 78.4 | 28.5 | 1.72 | |
| DREAM-RTarget=Q235B, Draft=Q2B2026.05 | 77.89 | 68.05 | 2.26 | |
| DREAM-RTarget=Q32B, Draft=Q2B2026.05 | 76.1 | 67.98 | 2.48 | |
| DREAM-R-NSTarget=Q235B, Draft=M7B-RL2026.05 | 76.1 | 25.56 | 1.56 | |
| DREAM-R-NSTarget=Q32B, Draft=Q2B2026.05 | 75.55 | 60.06 | 2.19 | |
| Standard SDTarget=Q235B, Draft=Q4B2026.05 | 75.36 | — | 1.1 | |
| DREAM-RTarget=Q32B, Draft=R4B2026.05 | 75.25 | 56.1 | 2.03 | |
| DREAM-RTarget=Q32B, Draft=M7B-RL2026.05 | 75.2 | 44 | 1.66 | |
| Standard SDTarget=Q235B, Draft=Q2B2026.05 | 75.17 | — | 1.16 | |
| DREAM-R-NSTarget=Q235B, Draft=Q2B2026.05 | 75.17 | 57.29 | 2.13 | |
| Standard SDTarget=Q235B, Draft=M7B-RL2026.05 | 75.1 | — | 1.07 | |
| LRTarget=Q235B, Draft=Q2B2026.05 | 74.51 | 52.12 | 1.61 | |
| Standard SDTarget=Q235B, Draft=R4B2026.05 | 74.2 | — | 1.09 | |
| Standard SDTarget=Q32B, Draft=Q4B2026.05 | 73.48 | — | 1.08 | |
| LRTarget=Q235B, Draft=M7B-RL2026.05 | 73.3 | 25.24 | 1.4 | |
| DREAM-R-NSTarget=Q32B, Draft=M7B-RL2026.05 | 73 | 41.5 | 1.52 | |
| LRTarget=Q32B, Draft=Q2B2026.05 | 72.97 | 66.85 | 2.13 | |
| DREAM-R-NSTarget=Q32B, Draft=R4B2026.05 | 72.8 | 50.9 | 1.93 | |
| DREAM-RTarget=Q235B, Draft=Q4B2026.05 | 72.53 | 47.22 | 1.53 | |
| LRTarget=Q32B, Draft=Q4B2026.05 | 72.12 | 49.29 | 1.66 | |
| Standard SDTarget=Q32B, Draft=R4B2026.05 | 72.1 | — | 1.09 | |
| DREAM-RTarget=Q235B, Draft=R4B2026.05 | 71.8 | 45.9 | 1.52 | |
| Standard SDTarget=Q32B, Draft=M7B-RL2026.05 | 71.1 | — | 1.03 | |
| LRTarget=Q32B, Draft=R4B2026.05 | 70.95 | 53.8 | 1.82 | |
| LRTarget=Q32B, Draft=M7B-RL2026.05 | 70.5 | 39.7 | 1.38 | |
| LRTarget=Q235B, Draft=Q4B2026.05 | 69.88 | 48.22 | 1.48 | |
| SpecReasonTarget=Q32B, Draft=Q4B2026.05 | 69.86 | 48.51 | 1.23 | |
| Standard SDTarget=Q32B, Draft=Q2B2026.05 | 69.72 | — | 1.12 | |
| LRTarget=Q235B, Draft=R4B2026.05 | 69.05 | 46.9 | 1.47 | |
| DREAM-R-NSTarget=Q235B, Draft=Q4B2026.05 | 68.52 | 44.23 | 1.51 | |
| DREAM-R-NSTarget=Q235B, Draft=R4B2026.05 | 67.95 | 42.7 | 1.49 | |
| SpecReasonTarget=Q235B, Draft=Q2B2026.05 | 67.87 | 68.75 | 1.3 | |
| SpecReasonTarget=Q32B, Draft=R4B2026.05 | 62.1 | 38.7 | 1.26 | |
| SpecReasonTarget=Q235B, Draft=R4B2026.05 | 57.1 | 35.6 | 1.21 | |
| SpecReasonTarget=Q235B, Draft=Q4B2026.05 | 55.43 | 36.88 | 1.23 | |
| SpecReasonTarget=Q32B, Draft=M7B-RL2026.05 | 51.6 | 23.4 | 0.83 | |
| SpecReasonTarget=Q32B, Draft=Q2B2026.05 | 39.8 | 21.35 | 1.41 | |
| SpecReasonTarget=Q235B, Draft=M7B-RL2026.05 | 11.3 | 24.79 | 1.09 |