Relevant only to Salesforce shops on Agentforce, and even then it is pilot-only with no public price and a US-first rollout.
Kenning's take
What it does
Koa is a reasoning model Salesforce made by post-training NVIDIA Nemotron 3 Super on synthetic CRM scenarios, so Agentforce agents pick the right actions in multi-step sales and service tasks. Salesforce hosts it inside its own infrastructure and says no customer data is used for training. It is offered inside Agentforce and the Data Cloud model catalogue, currently to select pilot customers.
- Post-trained from NVIDIA Nemotron 3 Super
- Selectable as a model provider org-wide in Agentforce
- Usable per agent or sub-agent in Agentforce builder
- Listed as a managed LLM in the Data Cloud generative models catalogue
- Hosted by Salesforce; no customer data used in training
Who it suits. Salesforce customers building Agentforce agents for sales, service and other CRM workflows.
Key facts
- Category
- AI models
- Made by
- Salesforce
- Based in
- United States
- Released
- 15 September 2026
- How to get it
- in Agentforce for select pilot customers; general availability expected in winter 2026 in U.S. regions
- Works on
- Salesforce Agentforce, Data Cloud
FAQ
When was Salesforce Koa announced?
Salesforce and NVIDIA announced Koa on 15 September 2026.
Can I use Koa now?
Only select pilot customers can use it in Agentforce. Salesforce expects general availability in U.S. regions in winter 2026.
Is Koa trained on customer data?
Salesforce says no. Its training data was built from synthetic CRM scenarios.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.