Physical Plausibility Verification on PhysicsIQ MAGI-1 24B V2V (N=16)
52.2Spatial IoUOracle (Upper bound)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Oracle (Upper bound)N=16, Generator=MAGI-1 24B (V2V)2026.06 | 52.2 | 36.3 | 39.3 | 0.002 | 72.9 | |
| GEOPHYSN=16, Generator=MAGI-1 24B (V2V)2026.06 | 47.2 | 33.6 | 34.6 | 0.003 | 64.5 | |
| WMRewardN=16, Generator=MAGI-1 24B (V2V)2026.06 | 43 | 31.4 | 31.2 | 0.003 | 62.29 | |
| Qwen3-VLN=16, Generator=MAGI-1 24B (V2V)2026.06 | 41.5 | 28.2 | 29.1 | 0.003 | 55.1 | |
| BaselineN=16, Generator=MAGI-1 24B (V2V)2026.06 | 41.3 | 26.5 | 28.8 | 0.003 | 50.01 | |
| Qwen2.5-VLN=16, Generator=MAGI-1 24B (V2V)2026.06 | 40.1 | 23.8 | 27.3 | 0.003 | 50.6 | |
| VideoMAEN=16, Generator=MAGI-1 24B (V2V)2026.06 | 39.7 | 27 | 27.6 | 0.003 | 52.6 |