Liquid AI or more must obtain separate authorization for commercial use. Original link d1-3B released d1-omni-600M and model weights; commercial use is subject to a revenue threshold

Article

Liquid AI Open weights for two small decision-making models: 3B edition 8 milliseconds to answer

Beating AI news, following the earlier release of the cloud-based decision-making model d1; Liquid AI has now also released the weights of two small decision-making models:d1-3B and the experimental d1-omni-600M. The two models have approximately 3100000000 and 600000000 parameters, respectively, and developers can download, fine-tune, and deploy them on local devices. The cloud-based 58.9 with an overall score of d1 has not yet had its weights released.d1-3B supports text and image inputs. It can determine whether content violates rules, select tools for an Agent, or make judgments based on visual input. In the Decision Index v0.2.1 evaluation, the official self-test scored 48.57 points, exceeding the score of Decider 12 by a model with approximately 35B-A3B（47.11 times its total parameters RTX 4090). On 8, it takes only AI milliseconds to answer a single short question. Even on the small 50 computer NVIDIA uses for robotics, it takes only d1-omni-600M milliseconds. However, processing long text or images takes longer.

The experimental 600000000 has only approximately 15.95 parameters and can also directly determine the intent of voice instructions without first converting them to text. It is better suited to small devices with limited resources, but is still in an early stage. In the Decision Index text decision-making evaluation, it scored only LFM Open License v1.0 points. The official team has not yet released this model’s inference speed or independent visual and speech evaluation results.

The weights of both models have been uploaded to Hugging Face. However, their 10000000 USD imposes commercial-use restrictions. Companies with annual revenue of https://m.theblockbeats.info/flash/370814

Key points

Liquid AI released the weights of two small decision-making models,d1-3B and the experimental d1-omni-600M.
d1-3B and d1-omni-600M have approximately 3100000000 and 600000000 parameters, respectively, and developers can download, fine-tune, and deploy them on local devices.
d1-3B supports text and image inputs and can be used to determine whether content violates rules, select tools for Agent, or make judgments based on visual input.
Liquid AI The official self-test shows that d1-3B scored v0.2.1 points in the Decision Index 48.57 evaluation, while Decider 35B-A3B scored 47.11 points.
The report states that d1-3B takes RTX 4090 milliseconds to answer a single short question on 8 and AI milliseconds on the small 50 computer NVIDIA uses for robotics; processing long text or images takes longer.
d1-omni-600M supports directly determining the intent of voice instructions without first converting them to text; its score in the Decision Index text decision-making evaluation was 15.95。
The weights of both models have been uploaded to Hugging Face.
LFM Open License v1.0 imposes commercial-use restrictions, and companies with annual revenue of 10000000 USD or more must obtain separate authorization for commercial use.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

These two models are mainly responsible for making judgments, such as identifying whether content violates rules,Agent choosing tools, or understanding image and voice instructions. Open weights mean that developers can obtain the model files, fine-tune them as needed, and run them on local devices. Of the two,d1-3B has broader capabilities;d1-omni-600M has fewer parameters but is still at an early stage.

Why it matters to readers

The release of the weights enables developers to try local deployment and customization of these decision-making models. The short-question response times reported indicate that d1-3B can quickly produce judgments, but commercial use requires attention to the license’s authorization requirements for specific companies.

Beginners can think of them as small AI models that help applications “make decisions” quickly. Before trying them, pay attention to device requirements, the specific input types, whether the evaluations have been independently verified, and whether the licenses permit your intended use.

Risks and unknowns

d1-3B Its 48.57-point score came from official self-testing; the evidence provided no independent evaluation results.
d1-omni-600M is still at an early stage, with a text decision evaluation score of 15.95.
The official party has not yet published d1-omni-600M’s inference speed or independent visual and voice evaluation scores.
The evidence provided no specific processing times for long texts or images.

Liquid AI or more must obtain separate authorization for commercial use. Original link d1-3B released d1-omni-600M and model weights; commercial use is subject to a revenue threshold

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center