Liquid AI Releases d1 Decision Model For Edge Devices Under 4GB

Liquid AI announced its d1 decision model on Wednesday, September 30, 2026, from its headquarters in Boston, Massachusetts. Co-founder and CEO Ramin Hasani leads the company, and he’s positioning d1 around edge computing rather than large cloud infrastructure. One caveat up front: details about its memory target and hardware support come from leaks and supply-chain reports. None of it has been officially confirmed yet.

Liquid AI’s Decision Model Announcement

The release marks Liquid AI’s latest move to bring neural-network Intelligence closer to the devices that collect and act on data. The d1 model uses liquid neural networks developed from research at the Massachusetts Institute of Technology, which gives the announcement a direct link to one of the most important research centres in modern AI.

Liquid AI’s approach differs from the usual cloud-first model, where devices send information to remote servers for processing. A local system can reduce the need for constant connectivity. Its usefulness, though, will depend on software compatibility, power consumption and the accuracy of its outputs.

Hasani’s leadership matters because Liquid AI has built its identity around neural networks that can adapt over time instead of relying only on fixed computational patterns. The d1 release turns that research direction into a named product aimed at developers working with robots, sensors, industrial systems and other edge hardware.

Reported Hardware Requirements And Memory Target

The d1 model was reportedly designed for edge computing devices with a target memory footprint under 4GB of RAM. That figure would place it in a very different category from AI systems that require workstation-class hardware or large server clusters. Even so, Liquid AI hasn’t published enough confirmed performance data to establish a direct comparison.

The reported processing requirements support standard ARM-based processors and NVIDIA Jetson hardware. ARM chips power a wide range of embedded devices. Jetson platforms are commonly used in robotics, computer vision and autonomous-machine development. Support for both hardware groups would give d1 a practical route into existing development environments.

The memory target also separates d1 from high-memory local AI systems, including the 64GB RAM Model covered in earlier reporting. That comparison shouldn’t be read as a performance ranking: RAM capacity alone does not determine inference speed, accuracy or energy efficiency. For more detail, see VentureBeat AI.

MIT Research And Liquid Neural Networks

Liquid AI’s architecture traces back to liquid neural networks developed through research at MIT

FAQs