Benchmark
Physical AI Bench Conditional Generation Mask mIoU
Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests. Scores are percentages.
- Models measured
- 2
- Scores recorded
- 9
- Sources
- 1
- IQ weight
- No weight published
Score distribution
- 75.2%
- Mean
- 76%
- Median
- 2.4%
- Std dev
- 71.8%
- 10th percentile
- 77.7%
- 90th percentile
- 70.9%
- Min
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Cosmos 3 SuperBlur· Physical AI Bench data | 78.4% |
| 2 | Cosmos 3 SuperEdge· Physical AI Bench data | 77.7% |
| 3 | cosmos-transfer2-5-2bAll· Physical AI Bench data | 76.5% |
| 4 | Cosmos 3 SuperDepth· Physical AI Bench data | 76.2% |
| 5 | Cosmos 3 SuperSeg· Physical AI Bench data | 76% |
| 6 | cosmos-transfer2-5-2bBlur· Physical AI Bench data | 75.3% |
| 7 | cosmos-transfer2-5-2bEdge· Physical AI Bench data | 74.4% |
| 8 | cosmos-transfer2-5-2bDepth· Physical AI Bench data | 71.8% |
| 9 | cosmos-transfer2-5-2bSeg· Physical AI Bench data | 70.9% |
Data credits: Physical AI Bench (SHI Labs), MIT (licence)