Wednesday, Sep 16 | --:--
Back to home

Salesforce and NVIDIA Announce Koa, a CRM Reasoning Model on Nemotron

On September 15, 2026 at Dreamforce, Salesforce and NVIDIA launched Koa, Salesforce’s first CRM reasoning model for Agentforce, post-trained from Nemotron 3 Super on a synthetic 27-year CRM corpus. Salesforce keeps the weights. Pilot now; GA winter 2026 in US regions.

Tech Insights Reporter 4 min read San Francisco, CA
Cover illustration for Salesforce and NVIDIA Announce Koa, a CRM Reasoning Model on Nemotron

TLDR

Salesforce and NVIDIA on Tuesday, September 15, 2026 at Dreamforce announced Koa, Salesforce’s first CRM reasoning model for Agentforce, built by post-training NVIDIA Nemotron 3 Super on a proprietary synthetic dataset modeled on ~27 years of CRM deployments. No customer data in training or inference, Salesforce said; it controls the weights and runs post-training and inference inside its own trust boundary. Vendor CRM Benchmark (update opportunity, route a case, schedule a follow-up): Koa matches or exceeds leading general-purpose models with 3× fewer errors. Already running an employee Slack agent inside Salesforce. Customer pilots: Formula 1, UChicago Medicine, Baxter Credit Union, 1-800Accountant, Engine, Xero. Pilot now; GA winter 2026, US regions. Separate: Nemotron models into Missionforce (gov / air-gap); post-trained SKUs for select customers in October 2026.

What shipped

Item Status
Koa in Agentforce Select pilots now; GA winter 2026 US
Training SFT + GRPO via NeMo RL / Gym / AutoModel; 14+ industries, synthetic only
Missionforce Operations GA now (US); post-trained NVIDIA models October

Ranking: Salesforce CRM Benchmark only—no Arena / Artificial Analysis listing for a CRM-specialist SKU. Treat 3× fewer errors as vendor, not independent.

Product-line de-dupe: not the same-day Salesforce in Claude plugin (separate Anthropic article), not Aug 26 Claudeforce partnership announce, not generic Nemotron open weights. This is Salesforce owning a CRM model.

Why this story matters

Enterprises are post-training American open weights instead of sending CRM actions to Claude or GPT. Watch: whether 3× fewer errors survives a third-party CRM bench, and whether winter GA stays US-only.

Sources

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading