Shapelearn Qwen 3.8 27B (13.1 GB VRAM)

ShapeLearn released optimized GGUF quantizations of the Qwen 3.8 27B model, with the full versions outperforming their earlier "Lite" releases across multiple GPU benchmarks. The GPU-5 variant (13.1 GB VRAM) achieves 99.63% of the original BF16 model's performance while supporting both text and image inputs, with options for faster speculative decoding through MTP or DFlash2 methods. All five models in the release sit on the quality-speed efficiency frontier, with smaller variants like GPU-4 offering competitive performance at reduced memory requirements.