Vercel Adds Liquid AI’s d1.2 Models to AI Gateway

Liquid AI’s d1.2 and d1.2-ultra decision models are now available through Vercel’s AI Gateway, adding support for audio input and image bounding boxes. The models return structured answers for classification, scoring and object location instead of generated text.
- Liquid AI’s d1.2 and d1.2-ultra are available on Vercel AI Gateway.
- The models support audio input and image bounding box questions.
- d1.2 is described as faster and lower-cost, while d1.2-ultra is the most accurate.
- Decision requests appear in AI Gateway logs and count toward budgets.
Structured answers for decision tasks
According to Vercel, the models are designed for decision requests that produce structured results. Developers can ask Boolean questions that return yes or no probabilities, Choice questions that select a category, or Score questions that rate an input against a rubric.
Multiple questions can use the same input in a single request. This allows one piece of audio, an image or other input to support several decisions without requiring separate requests for each question.
Vercel describes d1.2 as the faster, lower-cost model, while d1.2-ultra is Liquid AI’s most accurate decision model. Developers can select them in AI Gateway with the model names liquid/d1.2 and liquid/d1.2-ultra.
Audio and image examples
Vercel’s AI SDK example uses liquid/d1.2-ultra to classify a voicemail. The audio file is passed as a file part, and the request asks whether the caller is asking for a refund. The response exposes the result as a probability. The example requires ai 7.0.136 or later and uses the experimental_decide function.
For image classification, Vercel says developers can pass image bytes with the matching media type, such as image/png. A separate Python example uses the Decisions API to locate a red rectangle in an image. Box questions return the top and bottom as fractions of the image height, and the left and right as fractions of its width.
The examples require developers to set AI_GATEWAY_API_KEY to an AI Gateway API key before making requests. The image example sends its request to https://ai-gateway.vercel.sh/v1/decisions.
What the integration means for developers
The addition gives AI Gateway users a way to route decision-oriented tasks through Liquid AI’s two models, with interfaces for both the AI SDK and the Decisions API. The documented use cases include judging whether a voicemail contains a refund request and locating an object within an image.
These requests are visible in AI Gateway logs and count toward budgets, according to Vercel. Developers can consult the decision model documentation for more details or browse the available decision models.
The announcement expands the types of inputs supported by the integration beyond text-oriented workflows. Audio files can be used for Boolean decisions, while images can produce bounding box data in a structured response.


