Kosmos2: benchmark results
Provider: Microsoft. Released 2023-06-29. Access: Open.
Aggregate Rating 1154 ± 49, rank #9796 of 10552 rated models, from 454 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Gaze Estimation | 35 | Score (%) | 93 |
| CoreCognition Image + Video Conservation | 66.97 | Accuracy (%) | 91.6 |
| OpenVLM MMT-Bench - Font Recognition | 35 | Score (%) | 88.1 |
| OpenVLM MMT-Bench - Behavior Anomaly Detection | 60 | Score (%) | 84.5 |
| OpenVLM MMT-Bench - Color Constancy | 57.1 | Score (%) | 82.3 |
| OpenVLM MMT-Bench - Animal Keypoint Detection | 40 | Score (%) | 82 |
| OpenVLM COCO Captions | 14.5 | CIDEr | 81.9 |
| OpenVLM MMT-Bench - Jigsaw Puzzle Solving | 35 | Score (%) | 81.6 |
| OpenVLM MMT-Bench - Handwritten Retrieval | 35 | Score (%) | 81.1 |
| OpenVLM MMT-Bench - I2IT | 37.5 | Score (%) | 78.2 |
| OpenVLM MMT-Bench - Image Colorization | 40 | Score (%) | 73.1 |
| OpenVLM MMT-Bench - 3D Indoor Recognition | 35 | Score (%) | 68.2 |
Interactive version: theaggregate.ai/model?slug=kosmos2 · How It Works · Data refreshed daily, snapshot 2026-11-16.