Strata 0.1.45 is an updated open-source AI engine that allows users to run a 125-billion-parameter model offline on a gaming PC with at least 12 GB of VRAM, introducing image support and achieving up to 40% faster decoding speeds. The update enhances the smaller Qwen3.6-35B-A3B family with picture-reading capabilities, parallel requests for multiple conversations, and maintains byte-identical answers to its predecessor. While the new features significantly improve the engine's functionality, users should be aware of the high hardware requirements and potential trade-offs in quality when using aggressive quantization. The release is accompanied by detailed documentation on the changes, installation process, and caveats, aiming to facilitate user experience and transparency in the updates

Strata 0.1.45: Open-Source AI Engine Gets a Speed and Vision Boost

Strata 0.1.45, the MIT-licensed open-source AI engine shipped by maintainer Niko1221 today, lets you run a 125-billion-parameter model entirely offline on an ordinary gaming PC with 12 GB or more of VRAM. The release adds picture-reading vision to the smaller Qwen3.6-35B-A3B family, parallel requests for up to eight conversations, and decode speeds up to ~40% faster, while keeping flagship answers byte-identical to the previous build. The notes stand out for transparency, documenting exactly which speed tweaks got shipped and which were rejected or behind opt-in flags. Quality at aggressive quantization, high hardware requirements, and a lengthy first-run download remain the main trade-offs to keep in mind.

Strata 0.1.45: Open-Source AI Engine Gets a Speed and Vision Boost @ Linux Compatible