Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear
Salesforce is positioning Koa as an enterprise reasoning model that keeps sensitive work inside its own stack.
Built with Nvidia on the open-weight Nemotron base, Koa is trained for sales, marketing, and customer-support tasks rather than general frontier benchmarks. Salesforce says it used synthetic simulations, not customer data, to post-train the model. The pitch is lower token cost, stronger data controls, and automatic routing through Agentforce when a task needs reasoning. Salesforce is still working with outside labs, including Anthropic, but Koa reduces how often those models have to handle enterprise prompts. TechCrunch AI's note
Built with Nvidia on the open-weight Nemotron base, Koa is trained for sales, marketing, and customer-support tasks rather than general frontier benchmarks. Salesforce says it used synthetic simulations, not customer data, to post-train the model. The pitch is lower token cost, stronger data controls, and automatic routing through Agentforce when a task needs reasoning. Salesforce is still working with outside labs, including Anthropic, but Koa reduces how often those models have to handle enterprise prompts. TechCrunch AI's note
score 6