Decision models with calibrated probabilities
Grounded in available product and source data
Liquid d1.2 is a pair of decision models, d1.2 and d1.2-ultra, for tasks where a plain answer is not enough. They return calibrated probabilities, so a confidence score can be used to send the harder cases to a person.
This release adds audio input and five new task types: reading a value out of text, finding every span of a kind, locating objects in a picture, estimating a number with a range, and selecting several options at once. The models sit on the same API endpoint as earlier versions.