Published: 30 September 2026
Liquid AI released d1 on 29 September 2026, a model that classifies, routes and scores without generating a single output token. Instead of writing text, it returns a probability for every possible answer, so you are billed for input only. It runs through Liquid’s API and Vercel’s AI Gateway with a context window of 32,000 tokens.
What is the d1 decision model?
Most teams that use a language model for classification ask it to write a small piece of JSON. The model produces that text one token at a time and you pay for every one of them. d1 skips the writing step entirely. It reads your input and returns numbers.
It handles three kinds of question, and you can combine them in a single call:
- Noul: a yes or no question, answered with a probability between 0 and 1. Is this message spam? returns 0.92.
- Choice: a pick from named options, answered with a probability for each. Routing a support ticket might return billing 0.65, technical 0.30 and account 0.05.
- Score: a rating on an ordered scale, answered with a weighted position. An urgency score might come back as 1.85, sitting between medium and high.
Liquid points at content moderation, ticket triage, safety guardrails and approving an agent’s tool calls as the jobs it is built for. The documentation draws the line simply: if the answer is one of a known set of options, use a decision model, and if the answer is a new string the model has to compose, use a language model.
What you get and what you do not
- Licence: proprietary, hosted API only
- No downloadable weights and no GGUF, MLX or ONNX build, so you cannot run it on your own hardware
- Not fine-tunable on your own data
- Context window: 32,000 tokens
- Access: Liquid’s API, plus Vercel’s AI Gateway under the model ID liquid/d1, with OpenRouter planned but undated
- A free tier called d1:free, with paid rates unpublished
That last point is the one that blocks a business case. You cannot calculate the saving over your current setup, because Liquid has not said what d1 costs once you leave the free tier.
What Liquid AI has not published
Liquid AI wrote no launch post for d1. Its news page still ends at LFM2.5-VL-DSpark on 24 September 2026. Everything about d1 sits in the developer documentation, which describes the API and gives no accuracy figures, no latency figures and no price.
That is a real gap for a model whose entire pitch is calibration. Calibrated means the stated probability matches how often the answer turns out to be right, so a 0.9 should be correct roughly nine times in ten. Liquid asserts this and publishes nothing you can check it against. Its own guide tells you to measure calibration on your own traffic, which is honest advice and also an admission.
Secondary coverage has placed d1 at the top of the Jev Decision Index on Hugging Face, a leaderboard that scores decision models across 37 benchmarks. That index ranks open-weight models, and d1 has no weights. Until Liquid or the leaderboard’s maintainer explains how an API-only model was measured on it, that placement is not evidence you should act on.
How does d1 compare to an open alternative?
Fastino released GLiNER2.5-Decide on 25 September 2026, four days earlier, aimed at the same work. The difference is not the output format but where the model runs. Fastino ships weights you can host yourself. Liquid keeps d1 behind its API.
For a team already sending this traffic to a hosted model, that difference costs nothing and d1 is a straight swap. For a team classifying regulated or customer data inside its own network, it rules d1 out completely, and no price cut will change that.
What this means
d1 is worth a pilot, not a migration. Take the single classification route with the highest volume, run it against d1 and your current model for a week, and compare accuracy and cost side by side. That is about a day of engineering work and it answers the only question that matters for your data.
Do not plan anything wider yet. A model with no published price, no published accuracy and no launch post is not something to build a pipeline on. Liquid’s LFM line has shipped steadily through 2026, so the company is not the risk here. The missing numbers are. If your data cannot leave your own servers, skip d1 and look at an open-weight decision model instead. Worth testing, not yet worth committing to.
For more information, visit the official announcement of d1 in the Liquid AI documentation.
