Open-source engine runs 125B Qwen model on consumer GPUs at speed

A new open-source project called Strata makes it possible to run Qwen 3.8 Flash Next, a 125-billion-parameter model, on consumer-grade hardware like an RTX 4090 GPU. The tool offers one-click installation for Windows and Linux, generates output at 100 tokens per second, and exposes an API compatible with OpenAI and Anthropic interfaces on localhost. It also supports optional image inputs, expanding what's possible for local AI inference without enterprise hardware.

This development lowers the barrier for independent developers and small teams to experiment with large language models without cloud API costs or external dependencies.