
Apple Just Killed AI Subscriptions Forever (Mac Mini M5)
AI Master · 17:35
Deep technical comparison of Mac Mini M4 Pro for local LLM inference: 273 GB/s bandwidth (not 550), 7B-32B models run well at 20-30 tok/s, 70B barely usable at 3-5 tok/s, can't run GPT-OSS 120B at all (AMD Strix Halo wins there). Apple killed 64GB config and hiked prices $200 citing DRAM shortage. Bloomberg reports Apple skipping M6 Pro/Max/Ultra entirely, fast-tracking to M7 targeting 1.5TB unified memory and Nvidia Blackwell-class performance. Key MLX creator Aani Hanan left Apple for Anthropic along with ~12 AI researchers.
Alösha's take: The M7 roadmap leak and MLX talent exodus to Anthropic are the real signals here — useful context for anyone weighing local inference hardware timing.










