Salesforce and Nvidia introduce enterprise-focused reasoning model Koa
Salesforce has introduced Koa, its first reasoning model, developed with Nvidia from the open-weight Nemotron model. The companies post-trained it for sales, marketing and customer-support work, rather than primarily for general-purpose or advanced mathematical problems. Koa will be offered within Salesforce’s Agentforce platform, where companies create agents for routine tasks such as answering service questions and scheduling appointments.
Salesforce says Koa was trained without real customer data. The companies instead created synthetic scenarios that reproduced interactions involving customer-service and sales professionals, including difficult customer calls and deal-closing situations. The model is intended to perform relevant enterprise tasks with fewer tokens than sending comparable requests to models such as Claude or ChatGPT, potentially lowering usage costs.
Koa is also designed to operate within Salesforce’s data and security requirements and can be selected through Agentforce’s AI gateway, which routes requests to different models. Salesforce executive Jayesh Govindarajan said Nemotron provided a capable pre-trained base with clear data provenance. Salesforce is continuing to support other providers: its ClaudeForce partnership with Anthropic lets companies use Claude while keeping their data in Salesforce’s system.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.