Liquid AI's d1 decision model arrives on Vercel's AI Gateway
Liquid AI's d1 model is now available on Vercel's AI Gateway. According to Vercel, d1 is a decision model that evaluates shared state against typed questions and returns structured answers with probabilities, without generating text tokens.
- Liquid AI's d1 is now available on Vercel's AI Gateway.
- d1 is a decision model that returns structured answers with probabilities, without generating text tokens.
- It supports image input for visual classification, inspection and scoring.
- It can be used via the AI SDK decision API, the OpenAI-compatible Decisions API or the TypeSafe-compatible API.
- Decision calls appear in logs and count toward budgets alongside other model requests.
A model built for classification, routing and scoring
Vercel describes Liquid d1 as a decision model. It takes a shared state and a set of typed questions, then evaluates the state against those questions. The stated use cases are classification, routing and scoring.
The key difference from a typical language model is the output. According to Vercel, d1 returns structured answers with probabilities and does not generate text tokens. Developers get a defined answer to each question rather than free-form text that has to be parsed.
The model also supports vision. Vercel says d1 can answer typed questions about images, which covers visual classification, inspection and scoring.
Three ways to call the model
Developers can use liquid/d1 through three interfaces, according to Vercel:
- the AI SDK decision API
- the OpenAI-compatible Decisions API
- the TypeSafe-compatible API
The first example in the changelog asks whether a support agent issued a refund. It uses the AI SDK function experimental_decide, imported from the ai package. The call passes the model name, a state string reading "The support agent issued a full refund to the customer.", and a question called refunded of type boolean with the instruction "Was a refund issued?". The answer is read from result.answers.refunded.
Image input example
A second example shows image input. The code reads a local PNG file named square.png, encodes it as base64 and sends it as part of the state together with the text "Which color is the square?". The question is of type choice, with the options red, blue and green. The result is read from result.answers.color.
The example shows that the state can mix text and images in a single call, and that the question types define the shape of the answer.
What it means for AI Gateway users
AI Gateway, according to Vercel, provides a unified way to access models and track usage and cost. Decision calls appear in logs and count toward budgets alongside other model requests. Teams that already use the gateway can therefore add decision calls without a separate system for monitoring or spending limits.
Vercel points to its decision documentation for Choice, Score and Boolean question types, and to a page listing all decision models available on AI Gateway. The changelog does not give pricing or performance figures for d1.
Vercel: Liquid AI's d1 is available on AI Gateway
