Standard One's open 3B and 8B models answer yes/no, choice and score questions, with SH variants handling several questions about one input in a joint pass.
The Apache-2.0 family serves those answers through a System One-shaped interface. Its v4 SH benchmark figures come from the maker's runs on the public Decision Index subset. They are not verified full-board scores. Standard One and TypeSafe's Jev are separate models.
How you can use it
The supplied local server lets you ask fixed-choice questions without sending the scenario to a hosted model. Pick an SH version for several questions about the same text. Follow the documented prompt format: a default system message changes the input read by the trained head, the model part that scores answers.