Salesforce says Koa "matches or exceeds" leading models with 3x fewer errors. Its own paper: Koa 0.86, Opus 4.8 0.87, GPT-5.5 0.90 on CRM Bench. Good enough and cheaper is the real pitch. Believe the paper, not the PR. #Salesforce #Agentforce #NVIDIA #LLM
zurl.co/u1Qqb
