Open-Source Engine Brings 125-Billion-Parameter AI Model to Consumer Gaming GPUs

Strata, an MIT-licensed open-source inference engine, now enables users to run Qwen3.8-Flash-Next—a 125-billion-parameter model—on consumer-grade gaming graphics cards with as little as 12 GB of VRAM. The achievement addresses a fundamental barrier to local AI deployment: most modern large language models require datacenter-scale infrastructure or expensive cloud APIs. With Strata, developers and researchers can now operate sophisticated AI capabilities entirely on their machines, offline and private.