Liquid AI has released d1.2 and d1.2-ultra, MarkTechPost reports. These are decision models, not chat models: they return calibrated probabilities over a fixed set of answers and generate no output tokens, so only the input is billed. The new family accepts text, images, and audio, and adds five typed answer styles, including reading a value, finding spans, locating objects, estimating a number with a range, and picking every option that applies. Liquid AI says d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, the top score among decision models on that index, with d1.2 at 69. Those numbers are the company’s, on that index. The models run on Liquid’s hosted API only. No weights have been released; the company says it will open them “at some point.” Earlier d1, d1.1, and d1-ultra are now deprecated. I wrote earlier when Ollama added Liquid AI’s LFM3 and d1 models; this is a newer hosted decision line, not a local-weights drop.
I have not called the API for this post.
My take
A model that answers a typed question with a probability, instead of writing a confident paragraph, is a shape I would like to see more of. I’ll be more interested when the weights are actually out, or when somebody besides the vendor has kicked the index numbers hard.
God Bless, and Use the Source!