---
format: "aidr-story-markdown/v1"
id: "f0ee393325e0fd9004d0c37a43f605289aed3b28ce51c818e135ea66ae1ab92c"
canonical_url: "https://aidr.today/f0ee3933?lang=en"
title: "Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-17T18:10:21.000Z"
category: "Chips"
topics: ["inference"]
source_urls: ["https://twitter.com/anemll/status/2111517143087067309","https://news.ycombinator.com/item?id=50133500"]
summary: "Qwen3.8-27B runs on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw) with 24K context and both K/V on int8, while the delta net stays FP16 and decoding peaks at 70 tok/s."
---

# Qwen3\.8\-27B on the iPhone 18 Neural Engine at 15–70 tok/s \(1\.5 bpw\)

> [Open the canonical story](<https://aidr.today/f0ee3933?lang=en>)

**Published:** 2026-10-17T18:10:21.000Z
**Category:** Chips
**Topics:** inference

## Summary

Qwen3\.8\-27B runs on the iPhone 18 Neural Engine at 15–70 tok/s \(1\.5 bpw\) with 24K context and both K/V on int8, while the delta net stays FP16 and decoding peaks at 70 tok/s\.

## Sources

- [Story source](<https://twitter.com/anemll/status/2111517143087067309>)
- [Discussion](<https://news.ycombinator.com/item?id=50133500>)