Liquid AI's d1.2-ultra tops the Decision Index, adds audio input
d1.2 and d1.2-ultra add audio alongside text and images, plus five new answer types including bounding boxes and multi-select. Liquid says d1.2-ultra now leads its own benchmark.
Liquid AI released d1.2 and d1.2-ultra on 16 November, new versions of its “decision model” line built to answer a single typed question (a classification, an extracted value, a count) in one forward pass rather than generating free text. The headline change is audio: alongside text and images, both models now take spoken input too, covering 43 languages on the FLEURS benchmark.
Five new answer types ship with the update: reading a value out of the input, extracting spans of text, locating objects in an image with bounding boxes, estimating a number with a range, and picking every option that applies with a probability attached to each. On Liquid’s own Artificial Analysis Decision Index v2.1, d1.2-ultra scores 75, two points ahead of the previous best decision model, with a calibration error of 0.8 across 13 benchmarks and 22,505 questions. Pricing runs from $0.02 per million input tokens for d1.2 to $0.042 for d1.2-ultra; both are live today on Liquid’s own API, with OpenRouter and Vercel access due “soon”.
Why it matters: if a pipeline is burning a full chat model’s tokens just to classify, route or extract a field, that’s a decision task, not a generation task, and a model built for it answers faster and cheaper. Audio support means you can now point one directly at a voice note or call recording instead of transcribing first.
The caveat: the benchmark win is measured against Liquid’s own Decision Index, a suite the company built and publishes itself, not an independent leaderboard.