Quasar

About Quasar

Silx AI's Quasar project is an open-weight foundation model series designed for long-context intelligence, capable of processing millions of tokens efficiently using a novel continuous-time attention architecture that scales linearly. Built and pretrained on Bittensor's decentralized compute network (subnet 24), Quasar leverages a distributed system where independent miners and validators collaboratively train and verify model checkpoints. All models are released with full open weights under Apache 2.0, enabling unrestricted access and deployment. The project aims to advance large-scale, frontier-class language models without reliance on centralized GPU clusters, making high-quality, long-context AI models accessible and verifiable.