Salesforce and Nvidia develop Koa reasoning model for enterprise tasks

Salesforce and Nvidia have developed Koa, Salesforce’s first reasoning model, for sales, marketing and customer-support tasks. TechCrunch reported the model will be offered through Salesforce’s Agentforce platform as an alternative to frontier models.

Before Koa, Agentforce could route complex, multi-step requests to models such as Anthropic’s Claude or OpenAI’s ChatGPT through Salesforce’s AI gateway. Koa is intended to handle the enterprise tasks Salesforce customers want agents to perform, including customer-service interactions and scheduling appointments.

Koa is built on Nvidia’s open-weight Nemotron model and was reportedly post-trained using synthetic data. The training scenarios simulated sales and customer-service environments, including interactions with irate customers and sales professionals attempting to close deals. Salesforce and Nvidia said they did not use actual Salesforce customer data.

Salesforce says Koa is designed to provide more efficient reasoning and lower token costs for relevant tasks than routing the same work to Claude or ChatGPT. Nvidia has attributed those potential gains to Nemotron’s inference architecture, while Salesforce has described the model as having clear data provenance and a US-based foundation.

The available material does not include independent benchmarks confirming Koa’s performance, efficiency or cost advantages. Salesforce is also continuing to support frontier models, including through its ClaudeForce partnership with Anthropic. The extent of Koa’s customer availability and its commercial pricing have not been specified.