MiniCPM4-8B: benchmark results
Provider: OpenBMB. Access: Open.
Aggregate Rating 1445 ± 32, rank #3984 of 10916 rated models, from 13 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CSSBench - Privacy Leakage | 7.33 | Attack success rate (%; share of obfuscated malicious privac | 77.8 |
| CSSBench - Public and Political Safety | 36.11 | Attack success rate (%; share of obfuscated malicious public | 38.9 |
| CSSBench - Fraud and Hate | 48.67 | Attack success rate (%; share of obfuscated malicious fraud | 33.3 |
| CSSBench - Health and Medical Misinformation | 21.33 | Attack success rate (%; share of obfuscated malicious health | 33.3 |
| SuperCLUE General (July 2025) - Agent | 39.55 | Score | 23.9 |
| SuperCLUE General (July 2025) - Code Generation | 42.77 | Score | 19.6 |
| SuperCLUE General (July 2025) - Hallucination Control | 61.05 | Score | 19.6 |
| SuperCLUE General (July 2025) - Overall | 26.74 | Score | 17.4 |
| SuperCLUE General (July 2025) - Math Reasoning | 7.63 | Score | 15.2 |
| SuperCLUE General (July 2025) - Science Reasoning | 4.07 | Score | 15.2 |
| CSSBench - Illegal Activities and Compliance | 43.62 | Attack success rate (%; share of obfuscated malicious illega | 11.1 |
| SuperCLUE General (July 2025) - Precise Instruction Following | 5.34 | Score | 8.7 |
Interactive version: theaggregate.ai/model?slug=minicpm4-8b · How It Works · Data refreshed daily, snapshot 2026-11-08.