Liquid AI Releases Open-Weight d1-3B and d1-omni-600M: Multimodal Decision Models With Zero Output Tokens

Liquid AI has released two new open-weight decision models, d1-3B and d1-omni-600M, designed for edge computing environments. These models are trained from distinct backbones, with d1-3B derived from Liquid AI’s latest vision-language model, LFM2.5-VL-3B. Both models are available on Hugging Face, with d1-omni-600M marked as experimental. The release aims to provide developers with lightweight, efficient decision-making tools that can run on resource-constrained devices.

The models are part of Liquid AI’s broader d1 decision model family, which includes variants optimized for different use cases. The d1-3B model scores 48.57 on the Decision Index 0.2.1, surpassing all 4B and 9B models and even outperforming the 35B-A3B Decider model. This performance highlights the efficiency of the models despite their smaller size. The decision index measures a model’s ability to make accurate, context-aware decisions across a range of tasks.

The d1-3B model is trained from a decoder-only vision-language model, which allows it to process and generate text based on visual input. This architecture makes it well-suited for applications requiring real-time decision-making, such as robotics and autonomous systems. The d1-omni-600M model, while smaller, is designed to handle a broader range of tasks, making it a versatile option for developers. Both models are available for immediate use, with the d1-omni-600M being an experimental release that may see further refinements.

The release of these models could shift the landscape for edge computing by providing more efficient alternatives to larger, more resource-intensive models. Developers may benefit from reduced latency and lower computational costs, but they may also face challenges related to model governance and vendor lock-in. The open-weight nature of the models could foster innovation but may also raise questions about long-term support and integration with existing frameworks.

The availability of these models on Hugging Face and their performance on the Decision Index 0.2.1 suggest a growing trend toward lightweight, high-performance models for edge and embedded systems. As these models gain adoption, they may influence the development of future AI applications that prioritize efficiency and real-time processing. The open-weight approach also aligns with broader industry efforts to democratize access to advanced AI capabilities.