Liquid AI Released Two Multimodal Decision Models
The startup launched new open-weight models designed for real-time tasks like routing and intent classification.
Updated on Oct. 7, 2026 in Artificial Intelligence
Live Poll
Would you use AI tools designed for direct decision-making rather than generative text responses?
Liquid AI has released two open-weight decision models, d1-3B and d1-omni-600M, which are designed for high-speed tasks such as moderation and intent classification. Unlike standard generative models, these systems output probability distributions over fixed options rather than generating text tokens.
Why it matters
By providing a specialized architecture for decision-making rather than text generation, these models offer high-efficiency solutions for automated system routing. The release provides developers with optimized tools for low-latency applications on standard hardware.
The d1-3B model features 3.12 billion parameters and supports text and image inputs, while the d1-omni-600M model contains 587 million parameters and handles text, images, or audio. The d1-3B model utilizes a 32,768 token context window and supports a maximum audio clip duration of 30 seconds.
The players
Liquid AI
This startup develops advanced artificial intelligence models and architectures.
Hugging Face
This platform serves as the central hub for hosting and sharing open-source machine learning models and datasets.
NVIDIA
This technology company produces the hardware stack, including the RTX 4090 GPU, required for operating these models.
The details
The models are built to return probability distributions in a single forward pass, ensuring speed for real-time industrial tasks. The d1-3B architecture combines LFM2.5-2.6B with a vision-language model, and both models are currently hosted as checkpoints on Hugging Face.
Timeline
Liquid AI released the new d1 decision models on October 7, 2026.
The Tech Race
This release reflects the broader industry shift toward specialized small-language models that favor computational efficiency over general-purpose text generation. It positions these tools against more resource-heavy LLMs that typically require significant cloud infrastructure to run effectively.
Developers and smaller businesses can integrate these models into their applications for free provided they maintain annual revenue under $10 million. These tools enable faster automated moderation or intent classification on existing hardware, reducing the need for expensive server clusters.
The takeaway
These models represent a pivot toward task-specific AI that prioritizes low-latency decisions over conversational capabilities. Users should focus on identifying specific routing or classification workflows where speed is more valuable than long-form content generation.
Further reading
For more background on the evolving landscape of specialized AI architectures, visit the Artificial Intelligence section.
Live Poll
Would you use AI tools designed for direct decision-making rather than generative text responses?
