Liquid AI's d1.2 and d1.2-ultra are now available on AI Gateway
Liquid's d1.2 models return structured decisions for audio and image inputs, including probabilities, category choices, scores, and bounding boxes. The faster, lower-cost d1.2 and more accurate d1.2-ultra offer a distinct option for classification tasks.
**d1.2** and **d1.2-ultra** are available on AI Gateway with audio input and bounding-box support. Their Decisions API returns Boolean probabilities, category choices, rubric scores, or object boxes; multiple questions can use one input.
Use these models when an agent needs a structured decision, such as classifying a voicemail or locating an object in an image, rather than generated prose. Vercel's audio example requires **AI SDK 7.0.136 or later**.
**d1.2** and **d1.2-ultra** are available on AI Gateway with audio input and bounding-box support. Their Decisions API returns Boolean probabilities, category choices, rubric scores, or object boxes; multiple questions can use one input. Use these models when an agent needs a structured decision, such as classifying a voicemail or locating an object in an image, rather than generated prose. Vercel's audio example requires **AI SDK 7.0.136 or later**. d1.2 is described as faster and lower-cost, while d1.2-ultra is described as more accurate. Decision requests appear in Gateway logs and count toward budgets; the material gives no comparative measurements.
This adds audio input and object bounding boxes to the Gateway’s structured-decision options, extending beyond the classification and scoring use cases described for Gemini Decision 1 and Decision-1.1. It also makes the tradeoff between d1.2 and d1.2-ultra explicit, but supplies no measurements to validate it; audio use requires AI SDK 7.0.136 or later.