Salesforce announced its first CRM reasoning model for Agentforce named ‘Koa’, built on NVIDIA Nemotron. The company revealed the news at Dreamforce, where CEO Marc Benioff discussed the partnership with President of NVIDIA Jensen Huang.
Koa already matches or does better than leading model performance on CRM actions with 3x fewer errors, Salesforce says, citing its CRM Benchmark. Nemotron-based models and accelerated computing are also being brought to Missionforce – Salesforce’s military project – Salesforce and NVIDIA aim to bring ‘mission-specific AI’ to private clouds, air-gapped networks, and other secure environments.
What Is Koa?
Koa, which was developed in collaboration with NVIDIA, was built by post-training Nemotron 3 Super with a synthetic dataset modeled on enterprise knowledge from 27 years of Salesforce CRM intelligence.
Salesforce controls the model weights and does post-training and inference completely within its own trust boundary, meaning customers get a specialized, Salesforce-hosted option for Agentforce use cases.
Salesforce says that no customer data was used to train the Koa reasoning model.
Instead, its training corpus was built from “synthetic scenarios” which reflect the reasoning, use of tools, and decision-making skills Agentforce agents perform across CRM workflows – like generating leads, qualifying opportunities, and resolving service cases, Salesforce says.
Instead of relying on generic content, those scenarios were made to simulate real-world enterprise workflows across more than 14 industries, according to Salesforce. These include financial services, manufacturing, travel, and healthcare – with each scenario pairing up a persona with a specific task, and mapping their actions, along with the tool calls an agent has to make to finish them.
Salesforce applied Supervised Fine-Tuning (SFT) and reinforcement learning with Group Relative Policy Optimization (GRPO) in order to post-train the model with NVIDIA NeMo RL, NeMo Gym, and NeMo AutoModel.
Salesforce says that, by training on a targeted set of enterprise tasks, the model “developed deeper expertise” and learned to produce the right answer, and to “take the right action, step by step, to reach a goal”.
NVIDIA Nemotron open models – AI models released with publicly accessible weights, data, or training recipes – gave Salesforce control to build CRM intelligence into the model itself.
Salesforce runs Koa within its own infrastructure, controlling the weights and making sure no customer data crosses the trust boundary during inference.
Marc Benioff, Chair and CEO, Salesforce, said: “The most valuable thing Salesforce has built isn’t our platform – it’s the accumulated knowledge of how enterprise business actually works. With Koa, the knowledge is put inside the model itself.
“We trained a reasoning engine that understands the structure of a deal, the lifecycle of a service case, and the workflows that vary across industries. That’s a different kind of intelligence, and it runs entirely inside your trust boundary.”
Jensen Huang, founder and CEO of NVIDIA, said: “AI is creating a much larger opportunity for software. Every company needs useful AI, tailored to its knowledge, expertise, and work.
“NVIDIA Nemotron open models give Salesforce the foundation to turn decades of enterprise expertise into specialized AI with Koa, creating a CRM model that can reason and securely take action.”
Koa is available to select pilot customers now in Agentforce, and expected to be generally available this winter in U.S. regions.
Missionforce Operations is generally available now in U.S. regions. Post-trained NVIDIA models will be made available to select customers in October 2026.
Summary
Salesforce and NVIDIA have announced Koa, Salesforce’s first CRM reasoning model for Agentforce. It’s built on NVIDIA Nemotron, with the intention of helping agents reason through complex, multi-step workflows, trained on synthetic datasets.
The collaboration extends NVIDIA open models into Missionforce, allowing government and highly regulated organizations to deploy mission-specific AI in secure environments.
