Salesforce Enters AI Battlefield with NVIDIA-Powered Vertical Model Koa
At its Dreamforce conference, Salesforce launched two seemingly contradictory products: Claudeforce and Koa. The former is a joint product with Anthropic that allows customers to call Salesforce directly in Claude to complete tasks, while the latter is a vertical model developed by Salesforce and NVIDIA to counter large models like Claude.
Koa's performance on CRM task processing lags behind top-tier models but outperforms GPT-4.1. Salesforce's internal tests show that Koa can already meet most of the needs of enterprises, making it ready for the pilot phase.
Salesforce launched Koa to address three key issues: the trust crisis of closed-source models, its dependence on cutting-edge large models, and the AI transformation of SaaS companies. By developing a vertical model, Salesforce can ensure that customer data is not transmitted to third-party vendors, reducing the risk of data leakage.
NVIDIA provided Koa with its Nemotron open-weight model and post-training toolchain, engineering collaboration, and other support. NVIDIA's CEO Jensen Huang personally endorsed Koa, stating that every software company will become an AI company.