Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear
techcrunch ·
Salesforce and Nvidia have launched Koa, a new reasoning model announced at Dreamforce that serves as an open-weight alternative to proprietary frontier AI systems. Built on Nvidia’s Nemotron base, Koa is specifically post-trained for sales, marketing, and customer support tasks using synthetic data rather than actual customer information. This approach addresses enterprise concerns regarding data privacy, security, and the high costs associated with token usage in closed models like Claude or ChatGPT. By offering a sovereign American pre-trained model with clear data provenance, Salesforce aims to reduce reliance on external AI providers for complex reasoning tasks. The model integrates into Salesforce’s Agentforce platform, allowing companies to handle multi-step workflows more efficiently and cost-effectively. Despite this new partnership, Salesforce maintains its relationship with Anthropic through the Claudeforce integration, ensuring data remains secured within its infrastructure while leveraging Claude’s capabilities.
techcrunch