Liquid AI Releases d1.2 and d1.2-ultra: Omnimodal Decision

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


By Asif Razzaq β€” November 17, 2026


Liquid AI has released d1.2 and d1.2-ultra, the newest members of its decision model family. Unlike conventional language models, a decision model does not generate free-form text; it returns calibrated probabilities over a fixed set of answers. The d1.2 family now accepts text, images, and audio, and adds five new answer types. d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, the highest score among decision models on that index.


Note: Liquid AI has not disclosed parameter counts or opened model weights. The company says weights will be released "at some point."

TL;DR


  • Input capacity: Up to 65,536 input tokens per question and 128 questions per call (parameters not disclosed).
  • Deployment: Hosted Liquid API only; no weights yet.
  • Benchmark scores: d1.2-ultra 75 and d1.2 69 on the AA Decision Index v2.1.
  • Calibration: 0.8-point calibration error for d1.2-ultra across 13 benchmarks (Liquid AI test).

Setting the Scene: Why Decision Models Matter in 2026


As agentic systems move from experimentation to production, the bottleneck has shifted from text generation to reliable decision-making. Enterprises increasingly need models that output calibrated probabilities over a bounded answer set rather than open-ended proseβ€”a requirement for routing, classification, risk scoring, and structured tool invocation. Liquid AI's d1.2 family is positioned squarely in this niche, and its multimodal expansion reflects a broader 2026 industry trend: decision models are becoming omnimodal by default.


What Is Liquid AI d1.2?


d1.2 is a hosted decision model that accepts text, images, and audio as input and produces calibrated probabilities over a fixed set of answers. Compared with its predecessors, the new family adds five new answer types, expanding the range of structured outputs it can reliably score. The model supports up to 65,536 input tokens per question and can process 128 questions per API call.


What Is d1.2-ultra?


d1.2-ultra is the larger variant in the family. It reaches a score of 75 on the Artificial Analysis Decision Index v2.1, the top result among decision models tracked on the index. Liquid AI reports a 0.8-point calibration error for d1.2-ultra across 13 benchmarks in its own testing.


Benchmark Performance


| Model | AA Decision Index v2.1 | Calibration Error (Liquid AI test) |

|---|---|---|

| d1.2-ultra | 75 | 0.8 points across 13 benchmarks |

| d1.2 | 69 | Not disclosed |


The AA Decision Index (Artificial Analysis Decision Index v2.1) evaluates how well models produce calibrated probability estimates over fixed answer sets rather than free-form text. d1.2-ultra's score of 75 places it at the top of the index's decision model category.


Deployment and Availability


The d1.2 and d1.2-ultra models are available through the hosted Liquid API. Liquid AI has not released model weights; the company has indicated it will open them "at some point." Parameter counts for both models have not been disclosed.

via MarkTechPost

Related