Salesforce launches Koa, its first CRM reasoning model built on NVIDIA Nemotron 3 Super
Salesforce has officially unveiled Koa, its first proprietary CRM reasoning model, developed in partnership with NVIDIA. The model is built by post-training NVIDIA’s Nemotron 3 Super on a synthetic dataset derived from nearly three decades of enterprise CRM data. This announcement coincided with a keynote appearance by NVIDIA CEO Jensen Huang at the Dreamforce conference, where he emphasized…
Key points
- Salesforce launched Koa, a CRM reasoning model built on NVIDIA Nemotron 3 Super using synthetic enterprise data.
- Koa runs on Salesforce’s own infrastructure using open weights, ensuring zero customer data is used in training or inference.
- Customer pilots start in October 2026 with firms like Formula 1 and Xero, with general availability expected in Winter 2026.
A key differentiator for Koa is its reliance on open-weight models, allowing Salesforce to fine-tune and run the system entirely within its own infrastructure. This ensures that no customer data is used during training or inference, addressing significant privacy concerns. According to Salesforce, Koa matches or exceeds leading model performance on CRM-specific tasks while reducing errors by a factor of three compared to general-purpose models.
The model is already active in internal workflows, powering an employee agent within Slack. Customer-facing pilots are scheduled to begin in October with organizations such as Formula 1, UChicago Medicine, and Xero. General availability for U.S. regions is expected in Winter 2026. Jensen Huang also used the platform to argue that AI safety is an engineering challenge that should not hinder innovation, asserting that companies can achieve both speed and security.
Model page: Koa →
The story so far
2 episodes →- Salesforce launches Koa, its first CRM reasoning model built on NVIDIA Nemotron 3 Superthis story
‘Now We Can Know Everything and Do Anything,’ Jensen Huang Says at Dreamforce
NVIDIA Blog · 15 September 2026
Loading the full article…
This text was published by NVIDIA Blog and written by Brian Caulfield. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
6sources- Salesforce issues revenue target of over $63 billion for fiscal 2030, beating estimatesPress · CNBC Technology ·
- AI progress and responsibility: Nvidia, Anthropic and OpenAI CEOs weigh in at Salesforce’s annual eventPress · siliconangle.com ·
- Salesforce bets CRM experience can give its AI an edgePress · MarTech ·
- Salesforce Debuts Koa Reasoning Model for Agentforce, Trained on NemotronPress · Unite.AI ·
- Salesforce and Nvidia’s new reasoning model is everything the AI labs should fearPress · TechCrunch AI ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Anthropic removes Opus 4.8 fast mode on November 30 · 2 src
- Mistral releases Mistral Large 4 (Le Chonk) 1-trillion-parameter multimodal model on Hugging Face · 12 src
- Google announces Gemini 4 Carbon, its latest frontier AI model for cybersecurity · 7 src
- LiquidAI releases browser demo for LFM3-Omni voice AI · 1 src
- Datalab releases Surya 2.5 OCR model with 85.9% accuracy on OmniParseBench · 1 src
Comments
via GitHub Discussions