Moondream2: benchmark results
Provider: vikhyatk. Released 2024-03-14. Access: Open.
Aggregate Rating 1266 ± 43, rank #9114 of 10916 rated models, from 189 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMBench V1.1 EN - Image Scene | 89.6 | Accuracy (%) | 81.3 |
| OpenVLM MMBench V1.1 EN - Identity Reasoning | 97.4 | Accuracy (%) | 71 |
| OpenVLM MMBench V1.1 EN - Social Relation | 86.8 | Accuracy (%) | 70.8 |
| OpenVLM MME - Existence | 195 | Score | 63.5 |
| OpenVLM MMBench V1.1 EN - Image Topic | 97.8 | Accuracy (%) | 62.4 |
| OpenVLM AI2D - Atom Structure | 75 | Accuracy (%) | 61.7 |
| OpenVLM MMBench V1.1 EN - Action Recognition | 85.5 | Accuracy (%) | 52.7 |
| OpenVLM AI2D - Faults and Earthquakes | 64.3 | Accuracy (%) | 46.8 |
| OpenVLM MMMU - Accounting | 40 | Accuracy (%) | 46.1 |
| OpenVLM MMBench V1.0 EN - RR | 71.1 | Accuracy (%) | 45.4 |
| OpenVLM MM-Vet - Spatial Awareness | 50.7 | Score | 44.2 |
| OpenVLM MME - Scene | 156.2 | Score | 43.6 |
Interactive version: theaggregate.ai/model?slug=moondream2 · How It Works · Data refreshed daily, snapshot 2026-11-08.