Liquid AI Releases Open-Weight d1-3B and d1-omni-600M: Multimodal Decision Models With Zero Output Tokens
Zero output tokens. Liquid AI's new d1 models make real-time decisions in a single forward pass—no token streaming, no latency tax.
Why it matters
Liquid AI releases open-weight multimodal decision models (d1-3B, d1-omni-600M) that return typed, calibrated answers without generating text tokens. This challenges the token-streaming paradigm for latency-sensitive inference and offers a different model architecture for real-time classification and structured decision tasks.
The key facts
7 to knowTwo models released: d1-3B (text+image) and d1-omni-600M (text+image or text+audio)
Zero output tokens—answers returned in one forward pass with no text generation
Calibrated, typed outputs (classification/structured answers, not free text)
Target use case: real-time decisions with sub-token latency requirements
Open-weight release (weights available for deployment)
Multimodal architecture (vision and/or audio alongside text)
Decision-model family positioning (not general-purpose generation)
The story so far
Earlier coverage of this storyline
- Multimodal open d1 decision models for the edgeHugging Face Blog
- This story
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Liquid AI has released Open d1, two open-weight multimodal models in its d1 decision model family. d1-3B reads text and images. d1-omni-600M reads text with an image, or text with audio. Neither model writes text. Each returns calibrated, typed answers in one forward pass with zero output tokens.…