Cosmos-Reason2-32B: benchmark results
Provider: Other.
Aggregate Rating 1698 ± 41, rank #356 of 10916 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Physical AI Bench - Understanding Overall | 70.8 | Overall Score (%) | 100 |
| VANTAGE-Bench - Event Verification | 73.58 | Event Verification (%) | 83.3 |
| VANTAGE-Bench - Semantic | 71.94 | Semantic (%) | 83.3 |
| TAR-Bench - Temporal Description | 20.7 | Scaled BERTScore F1 (%; a description of what happens in a g | 80 |
| VANTAGE-Bench - Temporal | 40.25 | Temporal (%) | 70.6 |
| TAR-Bench - Binary QA | 67.7 | Accuracy (%; yes/no questions about the anomaly, one positiv | 70 |
| VANTAGE-Bench - Video QA | 70.29 | Video QA (%) | 66.7 |
| TAR-Bench | 36.7 | Mean score (%; unweighted mean of the ten task scores: accur | 60 |
| TAR-Bench - Causal Linkage | 26.5 | Scaled BERTScore F1 (%; an explanation of what caused the an | 60 |
| TAR-Bench - Multiple-Choice QA with Explanation | 53 | Scaled BERTScore F1 (%; a multiple-choice answer with an exp | 60 |
| TAR-Bench - Open-Ended QA | 32.5 | Scaled BERTScore F1 (%; free-form answers compared with the | 55 |
| TAR-Bench - Temporal Localization | 25.1 | Mean IoU (%; the predicted anomaly interval against the anno | 55 |
Interactive version: theaggregate.ai/model?slug=cosmos-reason2-32b · How It Works · Data refreshed daily, snapshot 2026-11-08.