Amidst the release of Fable 5.1 and Mythos 5.1, Leiolai 1 is breaking out of the data center paradigm by running a first-of-its-kind model designed to distribute inference across consumer hardware.
Most frontier models still arrive through data centers. More demand means more accelerators, more cooling, and more centralized capacity.
The scale of centralized computing is visible in NVIDIA’s $89 billion Data Center quarter.
Leiolai 1 was designed around a different premise: computation can come from devices people already own. A phone on a charger, a Mac sitting idle, or a PC with spare capacity can help run Leiolai. Apple’s M5 Ultra putting up to 512GB of unified memory on a desk reflects the broader shift toward edge computing.
Computation without a fixed ceiling
Leiolai is designed to grow and scale without a fixed ceiling. It can distribute inference across an open-ended set of devices, allowing inference-time compute to scale as more hardware becomes available.
Leiolai can add computation without adding another building full of servers.
Paying the people who provide the computation
Centralized providers pay chip companies, cloud operators, and data centers. Leiolai pays users who provide computation through hardware they already own.
The Leiolai app shows earnings and estimated water savings. Running inference on consumer hardware does not require data-center evaporative cooling.
Leiolai 1 is available now
Leiolai 1 has an 11-million-token context window and no response output limit. It is available in the Leiolai app on iPhone, iPad, Mac, Android, and Windows, and through the Leiolai API.
Fable 5.1 and Mythos 5.1 move centralized frontier models forward. Leiolai moves inference onto hardware people already own.
Sources Anthropic’s Claude Fable 5.1 announcement, Anthropic’s Claude Mythos 5.1 announcement, and Leiolai 1 launch details.