Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear

In a strategic move at the Dreamforce conference, Salesforce announced the launch of "Koa," its first reasoning model built on Nvidia’s open-weight Nemotron model. Koa has been specifically post-trained to excel in sales, marketing, and customer support tasks. Koa represents a shift in enterprise needs, moving away from the general-purpose models offered by frontier AI labs. Jayesh Govindarajan, EVP of Salesforce AI, explained that while the company previously relied on frontier model providers for reasoning tasks, they decided to build their own model to ensure sovereignty and transparency. Govindarajan noted that the previous challenge was the lack of a reliable, sovereign American pre-trained base model with clear data provenance. With Koa, Salesforce provides enterprise customers with a specialized alternative within its Agentforce platform, allowing companies to build AI agents capable of handling complex tasks without routing data to external frontier models.