A20 Pro runs Apple on-device model at 150 tok/s, half the TTFT of A19 Pro
rxwei · x · 2026-09-18
mweinbach benchmarked Apple's on-device foundation model on iPhone 17 Pro Max (A19 Pro) vs 18 Pro Max (A20 Pro). Long prompts: 2340ms TTFT/45 tok/s vs 1125ms/55; medium: 946ms/70 vs 437ms/71; short: 638ms/100 vs 354ms/150. A20 Pro roughly halves TTFT and boosts short-prompt throughput by 50%.
More from Infra
- Why 2027 is the year HDDs, not SSDs, win as generated AI media explodes storage demand — cocktailpeanut · 2026-09-18
- p0's Parallel Search lands on Baseten, bringing cheap accurate web search to open-weight models — baseten · 2026-09-18
- Neon backend goes GA with full primitives for apps and agents, branching-first design — matei_zaharia · 2026-09-18
- Ternary Bonsai 2: 27B Model Under 6GB Runs In-Browser on WebGPU, Keeps 98.2% Quality — xenovatech · 2026-09-18
- HyperQwen seeks 4090/5090 owners to push Qwen local inference speeds — iamMess · 2026-09-18
- REBCO superconducting tape: 50-tesla fusion magnets and the rare-earth angle — Anen-o-me · 2026-09-18