Salesforce and Nvidia Launch Koa, an Open-Weight Reasoning Model Built for Enterprise Work

Salesforce introduced Koa at its Dreamforce conference, the company's first reasoning model, built on Nvidia's open-weight Nemotron model. The two companies post-trained it together to handle sales, marketing, and customer-support tasks.

Koa reflects a widening split between what enterprises need from AI and what frontier labs offer. Salesforce says the model gives customers an open-weight alternative to closed systems, is trained for specific work rather than abstract problems, never ingested customer data, and uses fewer tokens to cut AI spending. It can be routed automatically through an AI gateway and follows customer data and security requirements inside Salesforce.

Koa joins the other models available in Agentforce, where customers build agents for routine work like answering service questions or scheduling appointments. Jayesh Govindarajan, EVP of Salesforce AI, told TechCrunch that reasoning had always depended on frontier providers "until now," and said the missing piece was a pre-trained base model with clear data provenance - something he said Nemotron provided.

Salesforce and Nvidia trained Koa on synthetic data mimicking customer patterns, not real customer data, simulating scenarios from irate callers to salespeople closing deals. Nvidia's Kari Ann Briski pointed to token efficiency as a key advantage. Salesforce also announced Claudeforce, a partnership with Anthropic letting companies use Claude as their AI interface while data stays in Salesforce's system of records.

Source: TechCrunch