Salesforce Unveils Koa Reasoning Model Amid Expanded Claude Partnership

Salesforce has introduced Koa, a reasoning model for sales and support agents built on Nvidia's open Nemotron 3 Super model. The company adapted this model with synthetic business scenarios across more than 14 industries without customer training data. This means that Koa can handle tasks such as updating deals and routing support tickets.

Salesforce is expanding its partnership with Claude, a large language model, by making it available to customers in multiple ways: as a customer, distribution partner, and competitor. The company's technical report describes turning Agent Script workflow rules into simulated tasks and rewarding successful actions through software tools.

Koa has been tested and scored 0.86 on CRM Bench, a test of customer relationship management tasks, which is close to the scores of Claude Opus 4.8 (0.87) and GPT-5.5 (0.90). Salesforce expects Koa's general availability in the US by winter 2026.

The company has also announced AIforce, which allows outside AI tools to access its data, business logic, permissions, and actions. This move extends the distribution role of Claude and gives Anthropic a direct role in Salesforce's own products.