Skip to content
TrustList
K
AI Model

Salesforce's CRM reasoning model, built for agents and run inside Salesforce's own infrastructure

About Koa

Salesforce's first CRM reasoning model

Koa is a reasoning model built for customer relationship work, unveiled by Salesforce and NVIDIA at Dreamforce on 15 September 2026, and designed to run Agentforce. The starting point was Nemotron 3 Super, an open NVIDIA model with 120 billion parameters and published training datasets, which Salesforce post-trained further using a proprietary synthetic dataset. Salesforce says that no customer data went into this training.

How it was trained

The material imitates business workflows in over 14 sectors, among them manufacturing, financial services, healthcare and travel. Every example sets a persona a task and lays out the series of tool calls an agent has to make to complete it. Post-training combined supervised fine-tuning with reinforcement learning, specifically Group Relative Policy Optimization, on NVIDIA NeMo tools, and a judge scored whether the simulated customer's problem was actually resolved.

What it does when no tool fits

Salesforce's engineers say they trained the model for the case that worries enterprises most: being asked for something that no available tool can do. In that case Koa is meant to say plainly what it cannot do and hand off, not call a tool that merely sounds similar.

Staying inside Salesforce

The weights stay with Salesforce, and both post-training and inference take place on its own infrastructure. As a result customer data never leaves its trust boundary, and traces from customers play no part in training. Koa already runs inside Salesforce, including in an agent in Slack, and its work includes changing an opportunity record, sending a case to the right queue and booking a follow-up.

Claims and availability

On Salesforce's own CRM benchmark, Salesforce says Koa does as well as the leading models on CRM actions, or better, with about a third as many errors. The comparison is relative, made against models Salesforce does not name and on tests it designed, and no absolute score has been published. Six pilot customers are named, from an accounting firm and a credit union to Formula 1, a university medical center and Xero. For now only chosen pilot customers can use Koa, and Salesforce expects wider release in US regions during winter 2026.

TrustList summary

Koa is Salesforce's CRM reasoning model, announced on 15 September 2026 at Dreamforce. Salesforce post-trained it from NVIDIA's Nemotron 3 Super on a synthetic dataset, and runs the model inside its own trust boundary so customer data does not leave it. It is available to selected pilot customers now.

Features

  • CRM actions: built for work such as changing an opportunity, sending a case to the right queue, or booking a follow-up.
  • Agent use: built for Agentforce, and already running inside Salesforce, including an agent in Slack.
  • Clear hand-off: trained to say plainly when no available tool can do a request and to hand off, rather than call a similar-sounding tool.
  • Training method: supervised fine-tuning and reinforcement learning on NVIDIA NeMo tools.

Who it serves

Salesforce names six pilot customers, including Xero, Formula 1, Engine and a credit union, in sectors that range from accounting to higher education and medicine.

Key facts

  • Announced: 15 September 2026 at Dreamforce
  • Availability: open only to chosen pilot customers at present, with broader release for US regions anticipated in winter 2026
  • Base model: NVIDIA Nemotron 3 Super

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

Koa launches

Trust Score

4/ 100
Trust Score: New

An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.

Verification
0/100 · 20%
Reviews
0/100 · 30%
Awards
0/100 · 15%
Transparency
18/100 · 20%
Engagement
0/100 · 15%
Joining soon
Recommendations
Coming soon
Complaints
Coming soon

Updated 9/18/2026

Request a demo or quote from Koa

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of Koa

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about Koa.

Koa alternatives

Similar listings buyers compare against this one.

Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio

Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure

Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0

Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions

Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection