DeepSeek-VL-7B-Chat: benchmark results
Provider: DeepSeek. Released 2024-03-11. Access: Open.
Aggregate Rating 1471 ± 27, rank #2822 of 10916 rated models, from 48 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMBench-Live - Attribute Reasoning | 94 | Accuracy (%; attribute reasoning questions; 5.9K newly gener | 100 |
| MMBench-Live - Coarse Perception | 97.37 | Accuracy (%; coarse perception questions; 5.9K newly generat | 100 |
| MMBench-Live - Fine-grained Perception (Single Instance) | 75.37 | Accuracy (%; single-instance fine-grained perception questio | 100 |
| MATRIX (Materials Science) - SEM-BSE Interpretation | 42 | Judge score (% of the maximum) on MATRIX sem-bse interpretat | 80 |
| MMBench-Live | 83.04 | Overall accuracy (%; multiple-choice questions over all six | 80 |
| MMBench-Live - Relation Reasoning | 85.6 | Accuracy (%; relation reasoning questions; 5.9K newly genera | 80 |
| Uni-SafeBench - Composite-Unsafe VQA | 90.8 | Both-Safe rate (%) on the 314 composite-unsafe visual questi | 73.1 |
| MATRIX (Materials Science) - Research Reasoning | 56.8 | Judge score (% of the maximum) on MATRIX research reasoning: | 71.4 |
| Uni-SafeBench - Dual-Unsafe VQA | 80.7 | Both-Safe rate (%) on the 455 dual-unsafe visual question an | 69.2 |
| Uni-SafeBench - Image-Unsafe VQA | 83.3 | Both-Safe rate (%) on the 467 image-unsafe visual question a | 69.2 |
| Uni-SafeBench - Malicious Text Generation | 97.8 | Both-Safe rate (%) on the 630 malicious text generation quer | 66.7 |
| Uni-SafeBench - Text-Unsafe VQA | 66.7 | Both-Safe rate (%) on the 454 text-unsafe visual question an | 61.5 |
Interactive version: theaggregate.ai/model?slug=deepseek-vl-7b-chat · How It Works · Data refreshed daily, snapshot 2026-11-08.