The Mac Studio M5 Ultra is about twice the M5 Max: twice the Blender score, 49 to 98 percent faster text generation in llama.cpp, and 1,200GB/s of memory bandwidth to the Max's 460. It also starts at more than twice the price, $5,499 against $2,499.

The base Ultra comes with 96GB of memory and a 1TB SSD, configurable to 512GB; the base Max with 36GB and 512GB, up to 128GB.

Mac Studio M5 Max and Mac Studio M5 Ultra

Mac Studio M5 MaxMac Studio M5 Ultra
Price$2,499Apple US price, 32-core GPU, 36GB / 512GB$5,499Apple US price, 30-core CPU, 64-core GPU, 96GB / 1TB
ChipApple M5 MaxApple M5 Ultra
CPU cores18 (6 super, 12 performance)30 (10 super, 20 performance); 36 optional
GPU cores32 (40 optional)64 (80 optional)
Memory (base)36 GB96 GB
Geekbench 6 CPUSingle-core, points4,3554,310†
Geekbench 6 CPUMulti-core, points30,38437,870†
Geekbench 6 GPUMetal, points238,904233,865†
Local LLM (llama.cpp / Ollama)tg128 Llama 2 7B Q4_0, tok/s120179
Local LLM (llama.cpp / Ollama)pp512 Llama 2 7B Q4_0, tok/s3,2204,945
Blender Benchmarksamples/min6,46213,502
BuyCheck price on AmazonCheck price at Best BuyCheck price on AmazonCheck price at Best Buy

Figures as published by Blender Open Data (Oct 19, 2026), Macworld (Sep 21, 2026), The Verge (Sep 21, 2026), llama.cpp community benchmarks (GitHub) (Oct 19, 2026). † From another publication than the rest of its row, so not a like-for-like test. The best figure is marked where a row is all one source.

Compare every benchmark side by side

The figures, one lab per row

Every row is one source's figures for both: Blender Open Data and llama.cpp's community benchmarks on Apple silicon.

Mac Studio M5 Max vs Mac Studio M5 Ultra, same test, same lab

5 rows
Test and settingMac Studio M5 MaxMac Studio M5 UltraAheadLab
llama.cpp, Llama 2 7B Q4_0, prompt processing (tok/s)3,2204,945Mac Studio M5 Ultra by 54%llama.cpp community benchmarks (GitHub)
llama.cpp, Llama 2 7B Q4_0, text generation (tok/s)119.9179.1Mac Studio M5 Ultra by 49%llama.cpp community benchmarks (GitHub)
llama.cpp, Llama 2 7B F16, text generation (tok/s)37.173.6Mac Studio M5 Ultra by 98%llama.cpp community benchmarks (GitHub)
Blender Benchmark (samples/min)6,46213,502Mac Studio M5 Ultra by 109%Blender Open Data
Blender Benchmark, CPU (samples/min)584.71,142Mac Studio M5 Ultra by 95%Blender Open Data

A row is here only when the same publication ran the same test on both. Launch prices are the makers' suggested prices; check the shop for today's.

Free to reuse with credit and a link to this page (CC BY 4.0).

Local AI

Text generation speed follows memory bandwidth, and the Ultra has 2.6 times the Max's. In llama.cpp with Llama 2 7B at 4-bit it generates 179 tokens a second to the Max's 120; at 16-bit, 73.6 to 37.1. Prompt processing is 54 percent faster. More important for big models is how much memory you can buy: up to 512GB on the Ultra against 128GB on the Max. A 70B model at 8-bit needs about 75GB, which fits either; a 400B-class model fits only the Ultra.

Rendering

Blender Open Data puts the Ultra's GPU at 13,502 samples a minute to the Max's 6,462, and its CPU at 1,142 to 585. For 3D work, the Ultra halves render times.

Which one to buy

Buy the M5 Max for photo and video editing, code, and models up to 70B with 128GB. It is the better value for almost everyone.

Buy the M5 Ultra if you render 3D for a living or want to run the largest open models locally: it is the only Mac with more than 128GB.

Questions

About twice as fast in Blender (GPU and CPU), and 49 to 98 percent faster text generation in llama.cpp.

The M5 Max Mac Studio up to 128GB, the M5 Ultra up to 512GB.

The M5 Ultra, for its 1,200GB/s bandwidth and up to 512GB. For models that fit in 128GB the M5 Max is the better value.