Shapelearn Qwen 3.8 27B (13.1 GB VRAM)

ShapeLearn released full Qwen 3.8 27B GGUFs on August 18, 2026, improving on the earlier ShapeLearn-Lite set. The new models achieve 99.63% of BF16 benchmark scores on GPU-5 (13.1 GB VRAM) and 98.72% on GPU-4 (11.0 GB VRAM), while Speculative Decoding with MTP or DFlash2 boosts throughput across all tested GPUs.