Rent Mac Studio M5 Ultra for Local AI & LLMs

Rent dedicated Mac Studio M5 Ultra hardware with 256 GB or 512 GB unified memory. Hosted in Prague with secure SSH and remote desktop access.

Availability is confirmed by email. A request is not a reservation.

Configurations

  • Mac Studio M5 Ultra 36-core CPU, 80-core GPU · 256 GB unified memory · 2 TB SSD €60 / 1 day
  • Mac Studio M5 Ultra 36-core CPU, 80-core GPU · 512 GB unified memory · 4 TB SSD €150 / 1 day

Choose your configuration

Mac Studio M5 Ultra

36-core CPU, 80-core GPU · 256 GB unified memory · 2 TB SSD

1 day
€60
7 days
€359
30 days
€1,399
Request this configuration

Mac Studio M5 Ultra

36-core CPU, 80-core GPU · 512 GB unified memory · 4 TB SSD

1 day
€150
7 days
€899
30 days
€3,499
Request this configuration

Published LLM measurements

Six model results measured on a Mac Studio M5 Ultra with 256 GB unified memory.

Read the original measurement (bohemiansilicon.com)

Scroll horizontally on mobile to view every column.

Published generation speed and peak memory by model and context length.
Model Architecture Model size Empty context (tok/s) 8k context (tok/s) 32k context (tok/s) Peak memory · empty (GB) Peak memory · 32k (GB)
Qwen3.5 9B Dense 6.0 GB 167 161 141 5.5 GB 7.4 GB
gpt-oss-20b MoE 12.1 GB 184 172 145 11.7 GB 12.8 GB
Qwen3.5 27B Dense 16.1 GB 55 53 48 15.3 GB 18.8 GB
Gemma 4 31B Dense 28.8 GB 31 27 21 26.6 GB 31.5 GB
Qwen3.5 35B-A3B MoE 20.4 GB 162 154 141 18.8 GB 20.6 GB
gpt-oss-120b MoE 63.4 GB 120 113 96 59.5 GB 61.0 GB

The test used a 4 TB SSD, not the offered 2 TB configuration. The 512 GB model has not been benchmarked. Results vary by workload and are not guaranteed.

Test conditions

Tested 15 October 2026 on Mac Studio M5 Ultra, 256 GB unified memory and 4 TB SSD; macOS 27.0.1, Python 3.12, MLX / mlx-lm 0.32. Models: Qwen3.5 MLX conversions, gpt-oss MXFP4 and Gemma 4 QAT 4-bit. Each run used 512 prompt tokens and generated 128 tokens at context lengths 0, 8,192 and 32,768. Empty-context results average two runs; longer contexts were measured once. Random-token workload; peak memory includes model weights and cache.

Request a non-binding rental quote

Choose a configuration, start date and number of whole days. We confirm availability by email; submitting this form does not reserve a device.

Estimated price: —