DeepSeek-VL2: benchmark results
Provider: DeepSeek. Released 2024-12-13. Access: Open.
Aggregate Rating 1478 ± 13, rank #2685 of 10916 rated models, from 410 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SPD-Faith Bench - Difference Quantity Recall | 68.3 | Share (%) of pairs for which the model reports exactly the t | 100 |
| OpenVLM MMBench V1.0 EN - CP | 88.4 | Accuracy (%) | 97.4 |
| OpenVLM MMBench V1.0 EN - AR | 87.5 | Accuracy (%) | 96.7 |
| OpenVLM MMBench V1.1 EN - Function Reasoning | 96.7 | Accuracy (%) | 96.3 |
| OpenVLM SEEDBench IMG - Instance Location | 73.4 | Accuracy (%) | 95.8 |
| OpenVLM MMBench V1.1 EN - Future Prediction | 85.3 | Accuracy (%) | 95.2 |
| OpenVLM MMBench V1.1 EN - Image Style | 85.9 | Accuracy (%) | 95.1 |
| UltraVR - Remote Sensing | 44.2 | Accuracy (%) on the remote sensing domain (comparison, logic | 95 |
| OpenVLM MME - Celebrity | 172.4 | Score | 94.4 |
| OpenVLM MMBench V1.0 CN - CP | 88.2 | Accuracy (%) | 94.3 |
| OpenVLM MMBench V1.1 EN - Attribute Recognition | 96.6 | Accuracy (%) | 93.6 |
| OpenVLM MMMU - Biology | 60 | Accuracy (%) | 93.5 |
Interactive version: theaggregate.ai/model?slug=deepseek-vl2 · How It Works · Data refreshed daily, snapshot 2026-11-08.