Chakkde // AI News

d3 in Ollama: open decision models that score answers instead of chatting

Ollama 0.40.11 added d3, a family of decision models from vLLM Semantic Router. d3 doesn't write a chat reply. Give it text, structured state such as JSON, or an image, then ask a question: it picks a category, answers yes or no, or scores options, and returns a probability for each answer. One use is triaging support requests. Ollama offers 2B, 4B, 9B and 27B sizes; the 2B is about a 3 GB download, and the default d3 is the 4B (about 5.8 GB). The models are Apache 2.0, and Ollama's d3 packages use a 4,096-token context. The catch: the scores are model outputs, not guarantees, so Ollama advises checking them on examples from your own workflow before setting thresholds.

Try it

ollama run d3 "A customer says their card was charged twice." --ask "Is this a billing issue?"

Needs Ollama 0.40.11 or later. Swap in your own text and question; d3 returns a probability for each answer.

How we verify

Every claim in this story was checked against at least two independent, reliable sources, and quotes against their primary source. Anything we couldn't verify was left out. Found an error? Email corrections@chakkde.com.

Our full verification policy