Liquid AI Releases d1.2 and d1.2-ultra: Omnimodal Decision Models That Top the AA Decision Index

MarkTechPost
Read full story· 5 min
Liquid AI launched d1.2 and d1.2-ultra, omnimodal decision models handling text, images, and audio inputs with new answer types. d1.2-ultra leads the Artificial Analysis Decision Index v2.1 with a top score of 75.

What happened

Liquid AI introduced two new decision models, d1.2 and d1.2-ultra, which process text, images, and audio inputs. The models were launched via the hosted Liquid API, with d1.2-ultra achieving the highest score on the Artificial Analysis Decision Index v2.1. Both models add five new types of answers and support up to 128 questions per API call. The previous versions, d1, d1.1, and d1-ultra, have been deprecated.

Key details

  • d1.2-ultra scored 75 on the Artificial Analysis Decision Index v2.1; d1.2 scored 69.
  • Input limit is 65,536 tokens per question and 128 questions per call.
  • Audio, text, and images are all accepted as input; audio questions can be spoken.
  • Five new answer types include value extraction, span finding, object location, ranged estimates, and multi-label choices.

Why it matters

Organizations using decision models now have access to a tool that can handle multiple input types and provide more answer formats, which can streamline workflows that require text, image, or audio analysis. Teams can benefit from improved calibration, with d1.2-ultra reporting a 0.8-point calibration error, making it easier to trust the model's confidence scores and route uncertain cases to humans. The hosted-only deployment and lack of released weights mean users are dependent on Liquid AI's infrastructure for now.


More in Machine Learning