
D1 is Liquid AI's first decision model: instead of writing text, it reads what you give it and returns a yes/no answer, a choice from a list, or a score, each with a probability attached. Where the company's LFM language models generate sentences, D1 is built for quick, repetitive judgments such as sorting incoming requests, sending them to the right place, filtering out harmful content, or approving whether an AI agent may use a tool. The company says it is more accurate on problems written in many languages and on long inputs, and that stray instructions hidden in the material it judges mislead it less. There is no charge for output tokens, so each decision costs very little, and developers can plug it into their own products through OpenRouter. It cannot write free-form text or hold a conversation, its size and training details have not been published, and it is not yet available as a model anyone can download and run themselves.
See each evaluation’s detail page for its source and last updateData sources & removal requests