Announcing Koa: Salesforce’s First CRM Reasoning Model, Built on NVIDIA Nemotron
Salesforce trained a reasoning model on 27 years of CRM data. Koa runs on NVIDIA Nemotron and ships for enterprise agents.
Why it matters
A major vendor (Salesforce) is releasing a domain-specific reasoning model built on open weights (Nemotron), signaling the shift from general LLMs to specialized frontier models for vertical AI. This is both a capability release and a statement about the lab race: enterprises are now building their own reasoning layers.
The key facts
7 to knowModel name: Koa
Training data: 27 years of Salesforce CRM intelligence
Base: NVIDIA Nemotron (open model)
Use case: enterprise agent reasoning for multistep CRM workflows
Deployment: Missionforce (government/regulated orgs, on-prem or controlled)
Date: September 15, 2026
Strategic angle: open-weight model adoption by enterprise software vendor
Go to the source
Salesforce Newsroomsalesforce.com
Publisher excerpt: Koa is trained with 27 years of Salesforce CRM intelligence to enable agents to reason through the complex, multistep tasks required for enterprise work The collaboration also brings NVIDIA open models and accelerated computing into Missionforce, giving government and regulated organizations…