Unsloth v0.1.804-beta Adds Qwen3.8-Flash-Next and GLM-5.3-Flash Support for Unsloth Desktop
Unsloth released v0.1.804 beta on 2026 08 27, enabling local inference of Qwen3.8 Flash Next and GLM 5.3 Flash models in Unsloth Desktop with 5x faster RAM off…
Published on MyPrivateClaw
Aug 28, 2026, 4:30 PM UTC
Coverage date
Aug 27, 2026
Last updated
Aug 28, 2026, 4:30 PM UTC
News summary
Unsloth v0.1.804 beta was released on 2026 08 27, adding support for running Qwen3.8 Flash Next and GLM 5.3 Flash models locally in Unsloth Desktop with 5x faster RAM offloading inference. This release is bounded to Unsloth Desktop application. Qwen3.8 Flash Next is a 125B parameter MoE multimodal model from Qwen that runs on at least 75GB RAM or unified memory with no GPU VRAM required and supports up to 262K context window. This is bounded to Unsloth Desktop inference via GGUF and requires specific llama.cpp PR 27742. GLM 5.3 Flash is Z.ai's new 320B multimodal model with only 18B parameters active at a time, requiring 102GB combined RAM + VRAM for the 1 bit Unsloth Dynamic GGUF, and supports up to 1M context. This is bounded to Unsloth Desktop inference via GGUF. Qwen3.8 Flash Next's 1 bit Unsloth Dynamic GGUF is 79% smaller than BF16 while retaining 80% top 1 accuracy, with the Ngra…