Qwen3.8-27B on the iPhone 18 Neural Engine at 15–70 tok/s (1.5 bpw)
Researchers measured the performance of the Qwen 3.8‑27B large language model running on the Neural Engine of Apple’s iPhone 18. The benchmark showed the model processing between 15 and 70 tokens per second, corresponding to roughly 1.5 bits per weight. The results illustrate the capabilities of on‑device inference for sizable transformer models using Apple’s dedicated AI hardware.