Qwen 3.8 27B Vision MLX Quants: Low-Bit Performance Benchmarks
Top-Eye-8104 · reddit · 2026-08-26
The team released various MLX quantizations of the Qwen 3.8 27B Vision model (ranging from 8bit to 3.23bpw) and benchmarked them against community versions like lm-studio and mlx-community. Tested on an H200 with a custom dataset, their 11.8 GB DWQ quant achieved 70.32% top-1 agreement, significantly outperforming others of similar size. This version runs on a 16GB MacBook with a wired limit tweak, utilizing distillation from a BF16 teacher model.
More from Infra
- MiniCPM-V 4.7 runs a 35B MoE multimodal model on phones with just 3GB memory — nevrekaraishwa2 · 2026-10-15
- Nvidia Is Now Worth More Than the Entire Indian Stock Market — theteknosaur · 2026-10-15
- NVIDIA joins EU's AI in Science Enablers, offering compute, training and cloud resources — nordicinst · 2026-10-15
- Open source's hardware trap: data centers crowd out consumer GPUs, local LLMs become a privilege — Dante_77A · 2026-10-15
- NVIDIA backs EU AI in Science push as 35 supercomputers to serve 3M researchers — NVIDIA Blog · 2026-10-15
- Jensen Huang: 2027 AI bottleneck is 35GW of power, revenue justifies the capex — BG2 Pod · 2026-10-15