Unsloth Adds Local Inference for Qwen3.8-Flash-Next and GLM-5.3-Flash

Unsloth released v0.1.804-beta, enabling local inference for two large multimodal reasoning models¹. Qwen3.8-Flash-Next, a 125B-parameter model described as an early preview of Qwen4's architecture, runs on 75GB RAM via a 1-bit Dynamic GGUF that is 79% smaller than BF16 while retaining 80% top-1 accuracy and supports 262K context. GLM-5.3-Flash, Z.ai's 320B-parameter mixture-of-experts model with 18B active parameters, runs on 102GB of combined RAM and VRAM with up to 1M context. The release also includes automatic GPU-RAM offloading, chat recovery after disconnects, and pre-load memory estimates.