Salesforce introduced Koa Decisions on November 6, a decision model built from Google's open-weight Gemma that accepts text, images and audio. It is the next version of HyperClassifier, the router already embedded in Agentforce, and it joins the Koa family alongside the separate Koa reasoning model Salesforce built with NVIDIA, covered here in September.

Instead of generating text token by token, the model reads a request plus a user-defined list of categories and returns a selection, a yes-or-no decision or a score in one step — often a single token. The team's research preprint reports 85.5% accuracy on a production intent-routing benchmark, responses in about half a second, and median speed up to roughly 7.5 times faster than the frontier models tested under identical serving conditions. Those are the authors' own measurements; the predecessor's production figure — about 26 milliseconds against 1,446 for a general frontier model — belongs to HyperClassifier, not to Koa Decisions.

Anyone can try the demo after registering for a free Trailhead account, including non-customers, and Salesforce says adding a category such as returns needs no retraining. The wider integrations — Salesforce flows, Apex, prompt templates and a developer API — are planned, not generally available, so the demo does not establish availability beyond Agentforce routing.