ByteShape released ShapeLearn GGUFs for Qwen 3.8 27B, with the 3.84 bpw (GPU‑5) model achieving 99.63% of BF16’s aggregate score across eight benchmarks and the 3.23 bpw (GPU‑4) model reaching 98.72%. These results represent average BF16‑normalized performance on instruct and thinking tasks, placing all five new models on the measured quality/speed‑bpw frontier across six GPUs. Lower bit‑per‑weight (BPW) configurations thus offer near‑full‑precision accuracy while improving inference efficiency.