Qwen3.8-27B runs on iPhone 18 Neural Engine at 1.5 bpw

An HN post reports Qwen3.8-27B running on the iPhone 18 Neural Engine at roughly 15–70 tokens per second after 1.5-bpw quantization. It highlights progress in on-device LLM inference, though hardware and quantization details need verification.

• Developer vibecodes GPU server into controllable heater with Qwen2026-10-18
• Users hope Qwen 4 models will think less2026-10-18
• Community tests full n-gram table for Qwen 3.8 Flash Next2026-10-18
• China proposes export controls on 23 AI technologies2026-10-16
• User Builds Three 4x R9700 Local AI Servers with 384GB Total VRAM2026-10-15