vecszn/polystrata is a local inference runner that drives large AI models on an ordinary PC. Qwen3.8-Flash-Next runs on the Strata engine, while GLM 5.3 Flash, DeepSeek V4 Flash, LFM2.5, Qwen3.6, Ornith 1.5, Gemma 4, Tiel Coder, and MiniCPM-V are supported by a second engine written for PolyStrata. The latest release shown on GitHub is v0.1.45-poly.6. The provided information lists nothing as changed in this release, so no changes can be confirmed.
According to the README, it supports 9 models through the two engines and lets you switch installed models while the server is running. It offers one-click installers for Windows and Linux, and it can open OpenAI- and Anthropic-compatible APIs locally so apps and coding agents can connect. The project's purpose is therefore not only to run models but also to provide a setup that plugs into existing tools. The repository is written in C++ and is released under the MIT license.
That said, because the Qwen3.8-Flash-Next path relies on the Strata engine, the novelty lies mainly in bundling multiple models together with the installation and connection methods. The second engine is experimental: the README states it has only been tested on one 32GB graphics card and one 8GB laptop. The throughput figures given are labels on a screenshot, and no reproduction procedure remains in the code. So rather than applying the performance numbers directly to a typical PC environment, it is better to look at the supported range and the level of verification together.