Liquid AI has released d1-ultra, its most capable decision model to date. The d1-ultra model can process text, images, or a combination of both, answering typed questions with a probability for each potential answer in a single forward pass without generating output tokens.
Key points
- Liquid AI released its d1-ultra multimodal decision model.
- d1-ultra processes text, images, or both to answer questions.
- The model scores 65 on the AA Decision Index.
- It is available via the Liquid API for $0.042 per million input tokens.
This functionality allows developers to perform classification, routing, and scoring tasks without incurring costs for generated text. The d1-ultra model is available via the Liquid API, priced at $0.042 per million input tokens.
The model leads all eight categories in Liquid's 112-evaluation suite, achieving a score of 65 on the Artificial Analysis Decision Index, which Liquid reports as the number one ranking. Specific details regarding parameters and context length have not been disclosed.
By the numbers
- AA Decision Index score
- 65
- API cost per million input tokens
- 0.042
Decision models are designed to answer bounded questions with probabilities rather than free text. Liquid's documentation outlines three question types: Noul (yes/no with probability), Choice (one label from a set with probabilities), and Score (probability-weighted position on a rubric).
Liquid recommends decision models for tasks such as routing, moderation, guardrails, triage, and as replacements for LLM-as-judge scenarios, differentiating them from language models used for summarization, chat, and code generation.
The d1-ultra model evaluates each question against the same input in one call. The Liquid AI team stated that d1-ultra utilizes a larger open-weight backbone than its predecessor, d1, combined with significant Liquid-specific customizations and a decision training pipeline, resulting in substantial gains on knowledge-based tasks.
Sources and references
Original report
Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index
MarkTechPost
Read on MarkTechPost (opens in a new tab)Linked in the original report
- Linkliquid.ai/blog/d1-ultra (opens in a new tab)
- Linkconsole.liquid.ai (opens in a new tab)
- Linkdocs.liquid.ai/lfm/models/decision-models (opens in a new tab)
- Linkdocs.liquid.ai/lfm/models/d1 (opens in a new tab)
- Linkartificialanalysis.ai/evaluations/artificial-analysis-decisio… (opens in a new tab)
- Linkartificialanalysis.ai/articles/artificial-analysis-decision-i… (opens in a new tab)
Ref 000MUPCV
Cite this
APA
Interestana AI Editorial. (2026, October 16). Liquid AI Releases d1-ultra Multimodal Decision Model. Interestana. https://interestana.com/articles/liquid-ai-releases-d1-ultra-a-multimodal-decision-model-that-scores-65-on-the-aa-decision-index-000mupcv
MLA
“Liquid AI Releases d1-ultra Multimodal Decision Model.” Interestana, 16 Oct. 2026, interestana.com/articles/liquid-ai-releases-d1-ultra-a-multimodal-decision-model-that-scores-65-on-the-aa-decision-index-000mupcv.
Chicago
Interestana AI Editorial. “Liquid AI Releases d1-ultra Multimodal Decision Model.” Interestana, October 16, 2026. https://interestana.com/articles/liquid-ai-releases-d1-ultra-a-multimodal-decision-model-that-scores-65-on-the-aa-decision-index-000mupcv.
BibTeX
@misc{interestana_000mupcv,
author = {{Interestana AI Editorial}},
title = {{Liquid AI Releases d1-ultra Multimodal Decision Model}},
howpublished = {Interestana},
year = {2026},
month = oct,
url = {https://interestana.com/articles/liquid-ai-releases-d1-ultra-a-multimodal-decision-model-that-scores-65-on-the-aa-decision-index-000mupcv},
note = {Summary of reporting by MarkTechPost, https://www.marktechpost.com/2026/10/15/liquid-ai-d1-ultra-decision-model/}
}Credit the source
Original reporting: MarkTechPost, “Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index”, https://www.marktechpost.com/2026/10/15/liquid-ai-d1-ultra-decision-model/(opens in a new tab)
Follow this story
Get Artificial Intelligence in your inbox — new stories in 40 words or fewer, once a week.
Subscribe