The quantized model Qwen-3.8-flash-next-heretic has been released

Developers presented Qwen-3.8-flash-next-heretic in the GGUF format with 3.78 bpw quantization. The model is based on the Qwen4Exp MoE architecture and optimized for operation on local devices via the llama.cpp library. The technical configuration includes 48 layers and 512 experts, providing high computational density with compressed weights. The model supports operation with a contextual window of up to 262,000 tokens, allowing for processing of voluminous documents and long data sequences. A key feature of the release is the specialized setup for processing the Japanese language and multimodal input data, including images. The use of the GGUF format allows the model to be run on consumer equipment with limited video memory without significant loss of generation accuracy. This release expands the capabilities of local use of the Qwen4Exp architecture, providing users with a tool for tasks that require deep understanding of context and work with visual data. Optimization for llama.cpp ensures compatibility with most modern systems for local LLM launch.