Liquid AI has introduced d1, which it calls its first “decision model,” in a post on X. Instead of writing essays or holding conversations, the model is meant to pick an answer from a fixed set, quickly, inside software.
The company says d1 is the first model to beat one called Jev on Hugging Face’s Decision Index. It also claims stronger results on multilingual tests, more resistance to prompt injection, and support for longer inputs. All of that is Liquid AI’s own account. The text available for this article included no score tables, and Jev’s maker was not identified.
Liquid AI pitches the model for jobs where the output is a label or a route: sorting support tickets, scoring leads, flagging content, or deciding which tool an agent should call next. The company says d1 is live on its own API and coming soon to OpenRouter, the model marketplace.
The bet is that a small specialist can handle these high-volume calls more cheaply than a general-purpose model. Teams paying frontier prices to classify text should test d1 against their own labeled data before trusting a vendor-picked index.
Reported from Liquid AI’s own thread on X, as archived by Thread Reader. The source page carries no date.
