CRMArena-Pro Benchmark Shows Top Agents' Success Rate Drops in Multi-Turn CRM Tasks

CRMArena-Pro, built inside Salesforce with 19 expert-validated tasks, shows leading agents drop from 58% single-turn success to 35% in multi-turn CRM settings with near-zero confidentiality awareness; its authors caution that read-only benchmark scores prove task fragments, not job competence.

2026-09-03 ~ 2026-09-03 · 4 related posts