Phone AI Index
Speed board
Each row is one device, one model file and one GPU setting. Speed comes from the Speed check: a fixed prompt, a fixed seed, a warm-up, then five timed replies. The value is the median of those five. The interval is a distribution-free 95% interval for the median of the submissions. Rows with fewer than three submissions are marked provisional. What a result contains.
iPhone
| Model | Quant | Size | Decode tok/s [95% CI] | Prefill tok/s | TTFT | n | GPU | Conditions |
|---|---|---|---|---|---|---|---|---|
| bartowski/ Llama- | Q4_K_M | 2.02 GB | 7.3 [no interval] | 55.4 | 3163 ms | 1 provisional | GPU | 100% on battery, llama.cpp b11118 |