MiniCPM5-2B is token-efficient for a reasoning model, Artificial Analysis says
Artificial Analysis reports 19k output tokens per Intelligence Index task for MiniCPM5-2B, compared with 56k for Ling 3.0 Tiny and 33k for Granite 4.2 8B.
TLDR
Artificial Analysis reports that MiniCPM5-2B uses 19k output tokens per task on its Intelligence Index, including 11k reasoning tokens. That compares with 56k output tokens for Ling 3.0 Tiny and 33k for Granite 4.2 8B. The organization says output token use matters for the on-device and edge deployments the 2.6B model targets.
Combined views
—
2 Sources, first seen ago
