MiniCPM-Llama3-V2.5: benchmark results
A vision-language model combining SigLIP-400M with Llama 3 8B Instruct. Its multimodal checkpoint is distinct from the language-only Llama release. Provider: OpenBMB. Released 2024-05-20. Access: Open.
Aggregate Rating 1416 ± 21, rank #5334 of 10916 rated models, from 463 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Industrial Produce Anomaly Detection | 90 | Score (%) | 98.3 |
| OpenVLM MMT-Bench - Multiple Image Captioning | 100 | Score (%) | 97.6 |
| OpenVLM MMT-Bench - Order Hallucination | 60 | Score (%) | 96.6 |
| OpenVLM MMT-Bench - Geometrical Relativity | 75 | Score (%) | 96.1 |
| OpenVLM MMT-Bench - MIA | 80 | Score (%) | 95.6 |
| OpenVLM MMT-Bench - Anatomy Identification | 80 | Score (%) | 95.1 |
| OpenVLM MMT-Bench - Social Relation Recognition | 70 | Score (%) | 94.4 |
| OpenVLM LLaVA-Bench - Conv | 96.6 | Score | 93.8 |
| OpenVLM MMT-Bench - Count | 57.2 | Score (%) | 93 |
| OpenVLM MMT-Bench - Tech Engineering | 50 | Score (%) | 93 |
| OpenVLM MMT-Bench - Spot The Diff | 85 | Score (%) | 92.7 |
| OpenVLM MMT-Bench - Counting By Reasoning | 90 | Score (%) | 92.2 |
Interactive version: theaggregate.ai/model?slug=minicpm-llama3-v2-5 · How It Works · Data refreshed daily, snapshot 2026-11-08.