Strata: Run a 125B-parameter AI model on a normal gaming PC (github.com)

Strata has introduced a groundbreaking capability that enables users to run a 125-billion-parameter AI model, specifically Qwen3.8-Flash-Next, directly on a standard gaming PC with just one NVIDIA graphics card (12-24 GB of VRAM) and 64 GB of RAM. This accessibility allows for faster AI model interaction, with response rates of 60-95 tokens per second, significantly outpacing human reading speeds. The open-source tool is designed to install effortlessly on both Windows and Linux platforms, requiring only a single click to get started, making advanced AI technology more approachable for enthusiasts and developers alike. This development is particularly noteworthy for the AI/ML community as it democratizes access to high-capacity models that previously necessitated servers equipped with extensive resources. Users can also optimize performance based on their hardware configurations and follow a simple calibration process to ensure maximum efficiency. The ability to run complex models with substantial context (up to 128K tokens) on consumer hardware reshapes the landscape of AI development, allowing for more experimentation and innovation by individual users, startups, and small research teams without requiring significant investment in specialized infrastructure.