Liquid d1

d1 is Liquid AI’s first decision model, announced on X on 29 September 2026. It follows the same request shape as Jev, so it is a System One model: you send a state, which is plain text or a JSON object, plus one or more typed questions. It returns one answer per question and generates no text. Liquid has not published the base model or the parameter count.

The three question types match Jev. A Noul returns one probability that the answer is yes. A Choice returns the chosen option, a probability for every option and a confidence value. A Score returns a value on a rubric you define, counted from 0, plus the probability of each level. The usage.output_tokens field in every response is 0. Liquid’s docs call d1 through POST https://api.liquid.ai/decisions/v1/systemone and use TypeSafe’s Python and TypeScript SDKs with base_url set to https://api.liquid.ai.

Liquid lists classification, routing, scoring, moderation, guardrails, reranking, LLM-as-judge checks and model routing. Those map to support inbox triage, intent and model routing, LLM guardrails and semantic reranking. Its docs pair the probabilities with thresholds, such as blocking above 0.8 and sending the middle band to a person, which is confidence-gated actions. The launch post says d1 wins on multilingual evals, resists prompt injection better and handles longer inputs more effectively than Jev. It gives no numbers for any of the three. The one chart it does publish is Liquid’s own reproduction of Hugging Face’s Decision Index 0.2.1. There d1 scores 58.9 overall against 57.9 for Jev 1.13. It leads on Arts (45.5 against 37.7), Language (67.6 against 62.0) and Retrieval (60.7 against 55.4). It trails on Tools (74.1 against 75.1) and Knowledge (43.3 against 51.3). The public Decision Index data, last generated on 28 September 2026, had no d1 entry when checked on 30 September, so nobody outside Liquid has reproduced these figures.

Liquid says to use a language model for free-form text, creative writing, multi-turn conversation, multi-step reasoning, open-ended questions, summaries and code. d1 cannot explain an answer. The gap in Knowledge suggests testing it before you route fact-heavy questions to it. Liquid gives no limits on options, levels, questions per call or request rate, so test your largest case.

Sign in at console.liquid.ai, create an API key that starts with liquid_, and call the model d1:free. Liquid does not label it beta or preview. It publishes no paid price. The launch post said d1 would reach OpenRouter soon, but OpenRouter’s list of decision models did not include it on 30 September 2026. Vercel’s AI Gateway already lists it as liquid/d1. There are no downloadable weights.

Question types Choice Score Noul Max Choice optionsNot documented Score levelsNot documented Questions per callNot documented Total context32,000 tokens State budgetNot documented Rate limitLiquid documents no rate limits for d1. EndpointPOST https://api.liquid.ai/decisions/v1/systemone SDKsPython: typesafe-sdkTypeScript: @typesafe-ai/sdk Liquid publishes no context length. Vercel's AI Gateway lists 32,000 tokens for liquid/d1, so treat that as Vercel's figure. Liquid's docs do not give a maximum number of Choice options, Score levels or questions per call. Their examples use up to 5 options, 4 levels and 3 questions.

d1:free, 29 Sep 2026, The only model id in Liquid's docs. Liquid's model library lists d1 as API only, with no GGUF, MLX or ONNX files, and not trainable. Vercel's AI Gateway lists it as liquid/d1. Release notes

What people use d1 for, one page per pattern. Workflow controlStarter

Send a support ticket to Jev once with every question attached. Category comes back as a selected label, severity and frustration as numbers on scales you wrote, refund intent as a probability. Your code reads those values and decides what happens to the ticket. Choice Score Noul Workflow controlIntermediate

One Jev call reads an incoming request and returns its intent as a label plus a difficulty rating on a scale you wrote. Your router reads both numbers and picks the handler: deterministic code, a cheap model, an expensive one, or a human queue. Choice Score Workflow controlIntermediate

Jev returns a confidence value from 0 to 1 alongside every Choice and Score answer. Your code treats it as a separate axis: act automatically when it's high, confirm or flag when it's middling, hand the decision to a person when it's low. Riskier actions get higher bars. Choice Score Safety and qualityIntermediate

Put one Jev request in front of an LLM and one behind it. Yes/no questions return the probability that each hazard holds, a Score rates how much harm complying would do, and your thresholds turn those numbers into pass, review, block, or a crisis path. Noul Score Retrieval and knowledgeIntermediate