normalaitech
Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)
Hacker News · 18 Oct 2026, 04:10 AEST
Qwen3.8-27B has been confirmed to run on the iPhone 18 Neural Engine at speeds of 15–70 tok/s. The system supports a 24K context with both K/V on int8 and maintains FP16 for the delta net.
Key facts
- Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)
- Confirmed 24K context on iPhone with both K/V on int8.
- Delta net stays FP16.
- Decoding also peaks at 70 tok/s with easy to predict context.
- Apple Neural Engine on A20 is fast.
Read aloud
Qwen3.eight-27B has been confirmed to run on the iPhone eighteen Neural Engine at speeds of fifteen–seventy tok/s. The system supports a 24K context with both K/V on int8 and maintains FP16 for the delta net. Qwen3.eight-27B on the iPhone eighteen Neural Engine at fifteen–seventy tok/s one point five bpw Confirmed 24K context on iPhone with both K/V on int8. Delta net stays FP16. Decoding also peaks at seventy tok/s with easy to predict context. Apple Neural Engine on A20 is fast.
Sources in this cluster
- Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)
Hacker News · 18 Oct 2026, 04:10 AEST