Liquid AI releases models that make decisions without generating a...

Liquid AI releases models that make decisions without generating a single token

Liquid AI has released its first decision model, d1, available via API as d1:free. The model takes a state as text or JSON and returns probabilities for predefined options in a single call, generating zero output tokens.

Compiled by Sergey KostenchukPublished 2026-09-29Updated 2026-09-30

2026-09-29 Coding HuggingFace

🛠 Liquid AI releases d1: makes decisions without generating a single token

The first decision model: input is a state — text or JSON, output is probabilities for predefined options, zero output tokens. Three types of questions — yes/no, choice with a confidence field, rating on a scale — can be mixed in a single request; calibration is the vendor's phrasing.

🌍 Classification, scoring, and moderation today are the work of LLMs that spend token generation and JSON parsing on answers from a known set. d1 removes output billing, latency is predictable, and disputed cases based on confidence go to a larger model or a human.

👤 Try for free: d1:free, key from console.liquid.ai, SDK typesafe-sdk (Python) and @typesafe-ai/sdk (TypeScript). Rule: answer from known options — decision model, new text — LLM.

Source 1: https://docs.liquid.ai/lfm/models/decision-models Source 2: https://x.com/liquidai/status/2105003472332693869

Sources