DeepSeek-VL2-Tiny: benchmark results
Provider: DeepSeek. Released 2024-12-13. Access: Open.
Aggregate Rating 1386 ± 16, rank #6669 of 10916 rated models, from 828 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MEGA-Bench Task - Paper Review Acceptance | 60 | Task Score (%) | 96.5 |
| MEGA-Bench Task - ASCII Art 30 | 35.7 | Task Score (%) | 93 |
| MEGA-Bench Task - Face Keypoint Detection | 84.9 | Task Score (%) | 91.9 |
| OpenVLM MMMU - Music | 43.3 | Accuracy (%) | 91.2 |
| MEGA-Bench Task - Clip Stable Diffusion Generate | 64.3 | Task Score (%) | 90.6 |
| MEGA-Bench Task - Face Attribute Edit | 57.1 | Task Score (%) | 90.6 |
| MEGA-Bench Task - Multi Contain Repeat | 6.7 | Task Score (%) | 90.6 |
| OpenVLM MME - Existence | 200 | Score | 89.7 |
| MEGA-Bench Task - AV Human Multiview Counting | 26.7 | Task Score (%) | 88.4 |
| MEGA-Bench Task - Autorater Motion Guided Editing | 21.4 | Task Score (%) | 88.4 |
| MEGA-Bench Task - Red Teaming Captcha | 11.1 | Task Score (%) | 87.2 |
| MEGA-Bench Task - Visual Prediction Rater Openable Part Segmentation | 28.6 | Task Score (%) | 87.2 |
Interactive version: theaggregate.ai/model?slug=deepseek-vl2-tiny · How It Works · Data refreshed daily, snapshot 2026-11-08.