Core AI model zoo

item iphone18pro-run1-fresh-nominal.json iphone18pro-run3-warm-serious.json iphone18pro-run4-bench-nominal.json m4max-release-run1-under-load.json m4max-release-run2.json
run / status / verdict 20260928-082149 / done / PASS 20260928-083607 / done / PASS 20260928-084755 / done / PASS mac-20260928-072428 / done / PASS mac-20260928-084407 / done / PASS
device iPhone19,2 iOS 27.0.0 (24A437), arch h19p iPhone19,2 iOS 27.0.0 (24A437), arch h19p iPhone19,2 iOS 27.0.0 (24A437), arch h19p arm64 macOS 27.0.0 (26A428), arch h16c arm64 macOS 27.0.0 (26A428), arch h16c
load1 both assets cold (s) / Core AI cache MB before → after 6.14 / 0 → 1274 1.22 / 1274 → 1274 1.17 / 1274 → 1274 5.39 / 0 → 1274 2.33 / 1274 → 1274
load1 peak footprint (MB) 591 562 565 851 765
warmup first synthesis (s wall / s audio, rtf) 10.33 / 2.51, 4.120 3.73 / 2.51, 1.489 - / -, - 7.55 / 2.46, 3.069 2.22 / 2.46, 0.902
load2 same process (s) 0.52 0.53 - 0.80 0.65
e2e fixtures / eos 18 / 18 18 / 18 None / None 18 / 18 18 / 18
e2e codes == Python engine: runs / prefix frames 0 / 385 of 1672 0 / 385 of 1672 None / None of None 10 / 1637 of 1688 10 / 1637 of 1688
e2e RTF median / p90 / overall 1.530 / 1.832 / 1.592 1.745 / 1.938 / 1.716 - / - / - 1.048 / 1.092 / 1.048 0.808 / 0.835 / 0.806
e2e frame ms median / prefill ms median / first audio s median 33.5 / 58 / 2.04 35.5 / 60 / 2.41 - / - / - 37.6 / 70 / 1.63 27.8 / 54 / 1.22
e2e audio s / wall s 77.6 / 123.6 77.6 / 133.3 - / - 78.4 / 82.2 78.4 / 63.2
e2e peak footprint (MB) 643 720 - 953 987
e2e thermal start → end nominal → nominal fair → serious - → - nominal → nominal nominal → nominal
bench fixture, waited s, thermal en_2, 0, fair en_2, 300, serious en_2, 145, nominal en_2, 0, nominal en_2, 0, nominal
bench RTF median / min / max 1.869 / 1.816 / 1.914 0.930 / 0.922 / 0.935 0.920 / 0.915 / 0.924 0.988 / 0.952 / 1.002 0.702 / 0.691 / 0.709
bench frame ms / first audio s 38.5 / 2.64 33.4 / 4.96 33.0 / 4.91 36.0 / 1.46 28.3 / 3.54