Cosmos3-Nano: benchmark results
Provider: Other.
Aggregate Rating 1750 ± 39, rank #181 of 10916 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| TAR-Bench | 44.1 | Mean score (%; unweighted mean of the ten task scores: accur | 90 |
| TAR-Bench - Binary QA | 75.8 | Accuracy (%; yes/no questions about the anomaly, one positiv | 90 |
| TAR-Bench - Binary QA with Explanation | 59.3 | Scaled BERTScore F1 (%; a yes/no answer with an explanation, | 90 |
| TAR-Bench - Causal Linkage | 30.5 | Scaled BERTScore F1 (%; an explanation of what caused the an | 90 |
| TAR-Bench - Multiple-Choice QA | 86.2 | Accuracy (%; multiple-choice questions about the anomaly; ze | 90 |
| TAR-Bench - Open-Ended QA | 38.7 | Scaled BERTScore F1 (%; free-form answers compared with the | 90 |
| TAR-Bench - Scene Description | 24.9 | Scaled BERTScore F1 (%; a physical description of the scene, | 90 |
| TAR-Bench - Temporal Description | 25.6 | Scaled BERTScore F1 (%; a description of what happens in a g | 90 |
| TAR-Bench - Video Summarization | 21 | Scaled BERTScore F1 (%; a summary of the events in the clip, | 90 |
| SpatialGen-Bench (Text Answering) | 60.04 | Overall Accuracy (%) | 80.8 |
| TAR-Bench - Temporal Localization | 26.5 | Mean IoU (%; the predicted anomaly interval against the anno | 80 |
| VANTAGE-Bench | 60.67 | Overall (%) | 70.6 |
Interactive version: theaggregate.ai/model?slug=cosmos3-nano · How It Works · Data refreshed daily, snapshot 2026-11-08.