MiniCPM-V-4.5-8B: benchmark results
Provider: OpenBMB. Access: Open.
Unified ELO 1546 ± 21, rank #603 of 1607 rated models, from 31 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Holtercare-Bench - Event Counting (Text) | 51.38 | Accuracy (%; multiple-choice closed QA on how many times a g | 100 |
| Holtercare-Bench - Heart Rate Extremum Timing (Video) | 39.17 | Accuracy (%; multiple-choice closed QA on the millisecond ti | 83.3 |
| Holtercare-Bench - Heart Rate Extremum Timing (Text) | 35.99 | Accuracy (%; multiple-choice closed QA on the millisecond ti | 75 |
| Holtercare-Bench - Event Timing (Video) | 37.5 | Accuracy (%; multiple-choice closed QA on the millisecond ti | 66.7 |
| RealStreamEval - Backward Tracing | 44.2 | Score (%; RealStreamEval on OVO-Bench tasks and questions: f | 66.7 |
| Molmo2-CapTest | 29.3 | Caption F1 (%; LLM-judged statement precision and recall aga | 60 |
| OCRBench v2 | 58.8 | Average (self-reported) | 53.8 |
| RealStreamEval - Forward Active Responding | 14.9 | Score (%; RealStreamEval on OVO-Bench tasks and questions: f | 53.3 |
| OCRBench-V2 (zh) | 58.8 | Score (self-reported) | 52 |
| Holtercare-Bench - Presence (Text) | 54.9 | Accuracy (%; multiple-choice closed QA on whether a given rh | 50 |
| SuperCLUE-Ondevice (September 2025) - GUI Agent | 73.96 | Score | 50 |
| VIABLE - Flawed-Answer Preference | 34.8 | Gold-pick rate (%; 33,000 gold-versus-failure-injected pairs | 50 |
Interactive version: theaggregate.ai/model?slug=minicpm-v-4-5-8b · How It Works · Data refreshed daily, snapshot 2026-09-29.