Liquid AI launches d1: a decision model offering calibrated probabilities without output tokens

Liquid AI has launched d1, a decision model designed for structured choices rather than text generation. It provides calibrated probabilities across a fixed set of outcomes in a single call without generating output tokens. Intended for tasks like classification, ticket routing, scoring, moderation, and LLM-as-judge checks, d1 is deployable via a hosted API under the name d1:free on the Liquid API. It is not trainable and lacks GGUF, MLX, or ONNX weights for self-hosting. The model returns typed answers from predefined options without text generation, with zero usage output tokens per response. Liquid AI recommends using d1 when answers are among N known options, while LLMs are better suited for tasks requiring text composition. The API endpoint is POST https://api.liquid.ai/decisions/v1/systemone, with keys obtained from console.liquid.ai. Liquid’s migration guide highlights differences between d1 and LLMs with structured output, such as the practicality of probabilities for setting thresholds and increased consistency in repeated evaluations. The road-decider cookbook demonstrates d1’s application in a pixel-art survival racer, where state design significantly impacts decision confidence. d1 is positioned in a growing category of non-generative decision models, with features compared across vendors as of September 29, 2026.