AIGenerated AI news, written by AI and explained simply. Models, technology, culture and the people behind it.
d1 is a family of decision models made by Liquid AI that read text and images and answer questions by outputting probabilities instead of writing out words.
2 stories, first covered on Oct 7, 2026.
d1 is a group of AI models created by Liquid AI that work differently from most language models. Instead of generating text word by word, they read input such as text or images and return answers as probabilities. Liquid AI released the first version of d1 as open-weight models, meaning developers can download and run them themselves. Variants include the d1-3B and the larger d1-ultra.
Most AI models produce output one token at a time, which uses a lot of computing power. d1 models return their answer in a single pass with zero output tokens. That makes them much faster and cheaper to run, especially for tasks like routing requests or making real-time decisions. Because they handle both text and images, they are multimodal. The models can run on everything from large servers to smaller edge chips, which is useful for putting AI close to where data is collected.
Liquid AI's d1-ultra has ranked number one on an AI benchmark, according to the company. The open-weight release lets anyone use or build on the models rather than relying on a closed API.
Liquid AI launched the original d1 in late September 2026, sharing it with the developer community. In early October, the company announced the d1-ultra version along with additional open-weight models. The reaction among developers on places like Hacker News has been positive but modest, with discussions focusing on the speed and cost benefits of models that do not generate long text outputs.
A normal language model writes out answers token by token, which takes time and computing power. d1 reads the input and returns a probability-based answer in one step, with no text output to generate. This makes it faster and cheaper to run for tasks that do not need long-form responses.
Yes. Liquid AI released the models as open-weight, so developers can download and run them. The models are designed to work on a range of hardware, from servers to smaller edge chips.
d1-ultra is the larger variant in the d1 family. It processes both text and images and has been reported to rank number one on an AI benchmark, according to Liquid AI. It shares the same probability-based output approach as the rest of the d1 line.
New modelsLiquid AI Unveils d1-ultra Decision Model Ranking 1 on AI BenchmarkHerald of AI
The new model reads text and images to answer questions with probabilities instead of generating text, cutting costs for developers.
How it worksLiquid AI Opens Up d1 Decision Models for Real-Time, Edge AI ChoicesHerald of AI
Two new open-weight multimodal decision models return answers with zero output tokens — built for speed and routing tasks on everything from servers to edge chips.