MiniCPM3-4B: benchmark results

Provider: OpenBMB. Access: Open.

Unified ELO 1442 ± 47, rank #1569 of 2656 rated models, from 24 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
SuperCLUE-DSPSafeBench - Overall77.64Score77.3
BFCL V4 - Web Search2Accuracy (%)36.6
BFCL V4 - Memory9.46Accuracy (%)35.4
SuperCLUE General (October 2024) - Liberal Arts69.87Score34.1
BFCL V4 - Irrelevance Detection73.71Accuracy (%)29.3
SuperCLUE General (October 2024) - Overall53.16Score26.8
SuperCLUE General (October 2024) - Hard26.56Score24.4
SuperCLUE General (October 2024) - Science63.04Score22
BFCL V4 - Multi Turn3.5Accuracy (%)18.3
BFCL V4 - Non-Live AST70.54Accuracy (%)18.3
SuperCLUE General (December 2024) - Liberal Arts73Score16.2
SuperCLUE General (December 2024) - Science45.9Score16.2

Interactive version: theaggregate.ai/model?slug=minicpm3-4b · How It Works · Data refreshed daily, snapshot 2026-09-19.