Halogen Flash Server

Halogen Flash Server is a community inference engine built for Qwen3.8-Flash-Next on AMD Strix Halo. Its model repository was created in September, and the developer published version 0.18.3 on October 14. The English changelog adds NPU embedding and image-editing routes, an optional streaming safety check, and fixes for queue handling and systems with a second AMD GPU. The Flash checkpoints themselves did not change in this version. Performance figures on the project page are developer measurements on specified hardware, not a general speed guarantee.