UC Berkeley Releases FreeToken: Running Frontier-Scale AI Models on Gaming PCs

Researchers from UC Berkeley and UT Austin have released FreeToken, an open-source inference engine that enables gaming PCs and consumer laptops to run massive 290-billion-parameter AI models at interactive speeds. Published in August 2026 under Apache 2.0 licensing, FreeToken represents a fundamental shift in how frontier-scale models can be deployed locally without datacenter infrastructure.