Mac Studio M5 Max vs Ultra for AI: Don't Waste $5,700

The cheapest M5 Ultra Mac Studio costs $400 more than the 128 GB M5 Max and has 32 GB less memory. And on the big models most people run, the $10,799 Ultra writes only about 5 tokens a second faster than the $5,099 Max. Here's every Mac Studio config, what each one runs, how fast, and the one I'd buy. Which Mac Studio should you buy for running AI models locally? I priced every M5 Max and M5 Ultra configuration, worked out which models each one can hold (memory), how fast each one writes (memory bandwidth), and checked that math against real reviews from T-Break, BGR and Tom's Hardware. Where nobody has published a test, the number is my own estimate and is labelled "Our math" on screen. Product photos: apple.com. The short version: for serious local AI, the Mac Studio M5 Max (40-core GPU) with 128 GB at $5,099 is the one to buy: it runs 120B-class models at about 75 tokens a second. The 64 GB Max ($3,499) is enough for 30-70B models. Skip the 96 GB M5 Ultra ($5,499) for AI, because it holds less than the cheaper Max. Buy the 256 GB Ultra ($10,799) only for models over 128 GB, long prompts all day, or image generation. The 512 GB Ultra arrives late October with no price yet: wait for real tests. Chapters 0:00 Mac Studio M5 Max vs Ultra for AI: Don't Waste $5,700 0:53 What does each step buy? 2:55 Will the model fit? 4:41 How fast does each one write? 6:30 Where did the gap go? 8:11 When is the Ultra worth it? 9:46 Does it pay for itself? 11:04 Which one is yours? Sources daringfireball.net: https://daringfireball.net/2026/08/co... apple.com: https://www.apple.com/mac-studio/specs/ macos-tahoe.com: https://macos-tahoe.com/blog/mac-unif... bgr.com: https://www.bgr.com/2264114/mac-studi... tbreak.com: https://tbreak.com/mac-studio-m5-max-... apple.com: https://www.apple.com/newsroom/2026/0... macrumors.com: https://www.macrumors.com/2026/08/25/... techrepublic.com: https://www.techrepublic.com/article/... modemguides.com: https://www.modemguides.com/blogs/ai-... tomshardware.com: https://www.tomshardware.com/desktops... morphllm.com: https://www.morphllm.com/chatgpt-vs-c... eia.gov: https://www.eia.gov/outlooks/steo/rep... Mac Studio M5 Max, Mac Studio M5 Ultra, Mac Studio for AI, M5 Max vs M5 Ultra, Mac Studio 512GB, Mac Studio 128GB, best Mac for local AI, local LLM, run AI locally, unified memory, memory bandwidth, tokens per second, LM Studio, Ollama, MLX, GPT-OSS 120B, Qwen, Llama 70B, mixture of experts #MacStudio #AI #Apple