MiniCPM-V-2.6: benchmark results
OpenBMB's 8B multimodal model (August 2024) built on SigLIP-400M and Qwen2-7B, handling single-image, multi-image and video understanding. Provider: OpenBMB. Released 2024-08-06. Access: Open.
Aggregate Rating 1483 ± 6, rank #827 of 1635 rated models, from 1325 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MARINER - VQA OCR-Integrated QA | 91.58 | Accuracy (%) on MARINER maritime visual question answering ( | 100 |
| MEGA-Bench Task - MuMA Theory Of Mind Social Goal | 53.3 | Task Score (%) | 100 |
| OpenVLM A-Bench - Technical Quality | 77.2 | Accuracy (%) | 100 |
| MEGA-Bench Task - VLN Hindi Next Step | 46.7 | Task Score (%) | 98.8 |
| OpenVLM MMT-Bench - Chart To Text | 100 | Score (%) | 98.8 |
| MEGA-Bench Task - Music Sheet Note Count | 17.6 | Task Score (%) | 97.7 |
| OpenVLM MMT-Bench - Traffic Sign Understanding | 95 | Score (%) | 97.3 |
| MEGA-Bench Task - Egocentric Analysis Single Image | 77.8 | Task Score (%) | 96.5 |
| MEGA-Bench Task - MFC Bench Check Clip Stable Diffusion Generate | 64.3 | Task Score (%) | 96.2 |
| OpenVLM MMT-Bench - Count | 58.1 | Score (%) | 95.9 |
| OpenVLM MMT-Bench - Interactive Segmentation | 71.4 | Score (%) | 95.4 |
| OpenVLM MME - OCR | 192.5 | Score | 95.1 |
Interactive version: theaggregate.ai/model?slug=minicpm-v-2-6 · How It Works · Data refreshed daily, snapshot 2026-10-15.