Open-Source Strata Brings 125-Billion-Parameter AI Model to Consumer Gaming PCs in Minutes

A solo developer has released Strata, an open-source inference engine that enables frontier-class AI model inference entirely on consumer gaming hardware with a single-click setup, capturing over 7,900 GitHub stars in less than two weeks. Strata runs Qwen 3.8 Flash Next, a 125-billion-parameter mixture-of-experts model, on compatible gaming PCs including NVIDIA RTX 20/30/40/50 series cards and AMD Radeon models, eliminating cloud dependency, subscription costs, and per-token charges. Created by developer Niko Veit and released September 24, the project provides a web interface and OpenAI-compatible API, enabling private chat, code generation, vision inputs, and agent integrations with zero account requirements.