Salesforce and Nvidia Unveil Koa: A Purpose-Built Enterprise Reasoning Model
At its flagship Dreamforce conference this week, Salesforce introduced Koa, the company's first reasoning model, built on Nvidia's open-weight Nemotron foundation. The two companies collaborated to post-train Koa specifically for sales, marketing, and customer-support workflows—a significant departure from the generalist frontier models that have dominated the AI landscape.
Koa is a striking example of how enterprise AI needs are diverging from what proprietary frontier labs are offering. While closed labs encourage enterprises to upload files, code, prompts, and feedback directly into their models and agents—often at substantial cost—Salesforce is taking the opposite approach: delivering an open-weight alternative tailored to specific business tasks.
What Koa Offers Enterprise Customers
Salesforce's new model provides several advantages that directly address enterprise pain points:
- An open-weight alternative to closed frontier models
- Task-specific training focused on real work rather than abstract problem-solving
- No ingestion of actual customer data, eliminating the risk of data leakage to third parties
- Lower AI spending by using fewer tokens to accomplish the same work
- Automatic routing through an AI gateway based on task requirements
- Full compliance with customer data requirements and security protocols embedded within Salesforce
Koa will be offered as an alternative to other models available in Salesforce's Agentforce platform, where customers build agents to handle routine tasks such as answering customer service inquiries or scheduling appointments.
From Frontier Dependency to Self-Sufficiency
"We've built many small task-specific language models, which are part of Agentforce's portfolio," Jayesh Govindarajan, EVP of Salesforce AI, told TechCrunch. "But reasoning has always been something that we've relied on the frontier model providers for. Until now."
Previously, when an agent needed to reason through a long-running or multi-step task, those prompts would be routed to a frontier model like Claude or ChatGPT through Agentforce's AI gateway—the system that determines which model handles which request.
"One of the reasons we hadn't done this before—train our own enterprise-grade frontier model—we always wanted to, but the challenge has always been the lack of a pre-trained base model to start with," Govindarajan explained. "Until Nemotron came along, there was no sovereign American pre-trained model that was available, that was state of the art, and that had clear data provenance. We have no idea what Qwen trains on," he added, referring to the popular Chinese open-weight model produced by Alibaba.
Synthetic Data, Real Results
Post-training a model like Koa means transforming it from a general-purpose system into one deeply versed in sales and customer support knowledge. Importantly, Salesforce and Nvidia accomplished this without using any actual customer data. Instead, they crafted synthetic data that mimicked customer patterns and scenarios.
"We actually simulated a customer service environment with a persona customer service professional, including irate customers that call into the customer service center, all the way to a sales professional who's trying to close a deal," Govindarajan described.
Koa is designed to outperform Claude or ChatGPT on the work tasks Salesforce customers want agents to handle—and to do so more cheaply in terms of tokens consumed.
Why This Matters in 2026
As enterprises increasingly demand AI that is secure, cost-effective, and purpose-built for their operations, models like Koa signal a broader shift. The frontier labs' approach—maximizing scale and general capability—is being challenged by a new wave of models that prioritize data sovereignty, task specialization, and operational efficiency. For Salesforce's customers, Koa represents a meaningful step toward AI that works the way their businesses actually operate.
via TechCrunch AI