CamThink

Sub-4B Models Challenge Agentic Tasks; Cost Advantage Still Unproven

CamThink tested six sub-4B models on agentic tasks, finding rising intelligence density, but the self-built benchmark cannot yet prove stable replacem