parallelquant
November 17, 2026 · MarkTechPost

Liquid AI releases d1.2 models that output calibrated probabilities

Liquid AI released d1.2 and d1.2-ultra, decision models that take text, images and audio and return calibrated probabilities with zero output tokens. d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, topping that benchmark.

Why it matters: Following Perplexity's open-weight Decider models, this points to an emerging category: models built to classify and decide rather than generate text. For routing, moderation and agent gating, returning a calibrated probability without generating tokens is cheaper and easier to threshold than parsing a chatbot's answer, and a dedicated benchmark means vendors are now competing on it.

Related updates