Local models
Rent a Mac Studio for local LLMs.
On Apple silicon the CPU and GPU share one pool of memory, so a large model can stay resident. Rent up to 512 GB of it by the hour or the month.
Why a Mac Studio for local models
A model runs well when its weights, its context and its cache all fit in memory with room to spare. On a Mac Studio that memory is unified: MLX and llama.cpp can use the whole pool, where a separate GPU card is limited to its own VRAM.
Because each booking is a whole machine, nothing else competes for that memory. The numbers you measure are the model on that Mac.
How much memory a model needs
Weights take about the parameter count times the bytes per parameter. Leave 20 percent on top for context and the runtime. As a rough guide:
- 8B parameters: about 5 GB at 4-bit, 10 GB at 8-bit, 19 GB at 16-bit
- 32B: about 19 GB at 4-bit, 38 GB at 8-bit, 77 GB at 16-bit
- 70B: about 42 GB at 4-bit, 84 GB at 8-bit, 168 GB at 16-bit
- 405B: about 243 GB at 4-bit, 486 GB at 8-bit
Which config to rent
Memory is the hard limit. If a model does not fit, a faster chip will not help. For models up to 70B, start with an M3 Ultra at 256 GB. Above 100B, look at the 512 GB configs, which leave room for weights, context and the KV cache.
The tool below checks any model size against every config in the fleet and shows what it costs per hour.
Running a model on the Mac you rent
Connect over SSH, then install your runtime. With MLX that is a Python virtual environment and pip install mlx-lm; with llama.cpp it is brew install llama.cpp. The local models guide has the commands.
Download weights once to one folder and point every tool at it. Copy results off before the term ends: the disk is erased afterwards. See transfer files.
Check a model against every config
Pick a model size and a precision. Each config shows whether it fits and what it costs per hour.
Model fit
Weights × bytes per parameter + 20%
Needs about42 GB
- M1 Max · 32 GBToo large
- M2 Max · 32 GBToo large
- M4 Max · 36 GBToo large
- M5 Max · 36 GBToo large
- M4 Max · 48 GBTight
- M1 Max · 64 GBFits
- M2 Max · 64 GBFits
- M1 Ultra · 64 GBFits
- M4 Max · 64 GBFits
- M2 Ultra · 64 GBFits
- M2 Max · 96 GBFits
- M3 Ultra · 96 GBFits
- M5 Ultra · 96 GBFits
- M1 Ultra · 128 GBFits
- M2 Ultra · 128 GBFits
- M4 Max · 128 GBFits
- M2 Ultra · 192 GBFits
- M3 Ultra · 256 GBFits
- M5 Ultra · 256 GBFits
- M3 Ultra · 512 GBFits
- M5 Ultra · 512 GBFits
Cheapest fit: M1 Max · 64 GB at $229.00/mo
Book itFits leaves a quarter of memory for macOS and your tools. Tight can work with a short context.
Every config and price
The whole list in one table: memory, SSD, cores, the hourly and the monthly rate, and how many are free right now.
Pre-order rentals
Reserved now, handed over when the Mac is racked. The lower price for waiting.
M5 Ultra · 96 GB
Pre-order rental2 TB SSD · 30-core CPU · 64-core GPU
$679.00/mo
Pre-order$1.25/hr· Hourly sold out5 rental slots openM5 Ultra · 256 GB
Pre-order rental4 TB SSD · 36-core CPU · 80-core GPU
$1,049.00/mo
Pre-order$2.00/hr· Hourly sold out2 rental slots openM5 Ultra · 512 GB
Pre-order rental8 TB SSD · 36-core CPU · 80-core GPU
$1,749.00/mo
Pre-order$3.40/hr· Hourly sold out1 rental slot openM5 Max · 36 GB
Pre-order rental512 GB SSD · 18-core CPU · 32-core GPU
$249.00/mo
Pre-order$0.50/hr· Hourly sold out5 rental slots open
| Chip | Memory | SSD | CPU | GPU | HourlySold out | Monthly | Free | |
|---|---|---|---|---|---|---|---|---|
| M5 Ultra | 96 GB | 2 TB | 30 | 64 | $679.00/mo | 5Pre-order rental | Pre-order | |
| M5 Ultra | 256 GB | 4 TB | 36 | 80 | $1,049.00/mo | 2Pre-order rental | Pre-order | |
| M5 Ultra | 512 GB | 8 TB | 36 | 80 | $1,749.00/mo | 1Pre-order rental | Pre-order | |
| M5 Max | 36 GB | 512 GB | 18 | 32 | $249.00/mo | 5Pre-order rental | Pre-order |
Ready now
In stock, one to three of each size. Priced above pre-order because you get it now. Email [email protected] before you check out and we tell you exactly how soon yours can be ready.
M4 Max · 36 GB
Available512 GB SSD · 14-core CPU · 32-core GPU
$239.00/mo
Rent$0.45/hr· Hourly sold out3 freeM4 Max · 48 GB
Available1 TB SSD · 16-core CPU · 40-core GPU
$309.00/mo
Rent$0.60/hr· Hourly sold out2 freeM4 Max · 64 GB
Available1 TB SSD · 16-core CPU · 40-core GPU
$389.00/mo
Rent$0.75/hr· Hourly sold out2 freeM4 Max · 128 GB
High demand2 TB SSD · 16-core CPU · 40-core GPU
$559.00/mo
Rent$1.05/hr· Hourly sold outOnly 1 leftM3 Ultra · 96 GB
High demand1 TB SSD · 28-core CPU · 60-core GPU
$479.00/mo
Rent$0.90/hr· Hourly sold outOnly 1 left, then 10 pre-order slotsM3 Ultra · 256 GB
High demand2 TB SSD · 32-core CPU · 80-core GPU
$819.00/mo
Rent$1.55/hr· Hourly sold outOnly 1 left, then 5 pre-order slotsM3 Ultra · 512 GB
High demand4 TB SSD · 32-core CPU · 80-core GPU
$1,399.00/mo
Rent$2.70/hr· Hourly sold outOnly 1 left, then 1 pre-order slotM2 Ultra · 64 GB
Available1 TB SSD · 24-core CPU · 60-core GPU
$439.00/mo
Rent$0.85/hr· Hourly sold out2 freeM2 Ultra · 128 GB
High demand2 TB SSD · 24-core CPU · 60-core GPU
$539.00/mo
Rent$1.05/hr· Hourly sold outOnly 1 leftM2 Ultra · 192 GB
High demand4 TB SSD · 24-core CPU · 76-core GPU
$689.00/mo
Rent$1.30/hr· Hourly sold outOnly 1 leftM2 Max · 32 GB
Available512 GB SSD · 12-core CPU · 30-core GPU
$199.00/mo
Rent$0.40/hr· Hourly sold out3 freeM2 Max · 64 GB
Available1 TB SSD · 12-core CPU · 38-core GPU
$279.00/mo
Rent$0.55/hr· Hourly sold out2 freeM2 Max · 96 GB
High demand2 TB SSD · 12-core CPU · 38-core GPU
$339.00/mo
Rent$0.65/hr· Hourly sold outOnly 1 leftM1 Ultra · 64 GB
Available1 TB SSD · 20-core CPU · 48-core GPU
$339.00/mo
Rent$0.65/hr· Hourly sold out2 freeM1 Ultra · 128 GB
High demand2 TB SSD · 20-core CPU · 64-core GPU
$459.00/mo
Rent$0.85/hr· Hourly sold outOnly 1 leftM1 Max · 32 GB
Available512 GB SSD · 10-core CPU · 24-core GPU
$159.00/mo
Rent$0.30/hr· Hourly sold out3 freeM1 Max · 64 GB
Available1 TB SSD · 10-core CPU · 32-core GPU
$229.00/mo
Rent$0.45/hr· Hourly sold out2 free
| Chip | Memory | SSD | CPU | GPU | HourlySold out | Monthly | Free | |
|---|---|---|---|---|---|---|---|---|
| M4 Max | 36 GB | 512 GB | 14 | 32 | $239.00/mo | 3Available | Rent | |
| M4 Max | 48 GB | 1 TB | 16 | 40 | $309.00/mo | 2Available | Rent | |
| M4 Max | 64 GB | 1 TB | 16 | 40 | $389.00/mo | 2Available | Rent | |
| M4 Max | 128 GB | 2 TB | 16 | 40 | $559.00/mo | 1High demand | Rent | |
| M3 Ultra | 96 GB | 1 TB | 28 | 60 | $479.00/mo | 1High demandthen 10 pre-order slots | Rent | |
| M3 Ultra | 256 GB | 2 TB | 32 | 80 | $819.00/mo | 1High demandthen 5 pre-order slots | Rent | |
| M3 Ultra | 512 GB | 4 TB | 32 | 80 | $1,399.00/mo | 1High demandthen 1 pre-order slot | Rent | |
| M2 Ultra | 64 GB | 1 TB | 24 | 60 | $439.00/mo | 2Available | Rent | |
| M2 Ultra | 128 GB | 2 TB | 24 | 60 | $539.00/mo | 1High demand | Rent | |
| M2 Ultra | 192 GB | 4 TB | 24 | 76 | $689.00/mo | 1High demand | Rent | |
| M2 Max | 32 GB | 512 GB | 12 | 30 | $199.00/mo | 3Available | Rent | |
| M2 Max | 64 GB | 1 TB | 12 | 38 | $279.00/mo | 2Available | Rent | |
| M2 Max | 96 GB | 2 TB | 12 | 38 | $339.00/mo | 1High demand | Rent | |
| M1 Ultra | 64 GB | 1 TB | 20 | 48 | $339.00/mo | 2Available | Rent | |
| M1 Ultra | 128 GB | 2 TB | 20 | 64 | $459.00/mo | 1High demand | Rent | |
| M1 Max | 32 GB | 512 GB | 10 | 24 | $159.00/mo | 3Available | Rent | |
| M1 Max | 64 GB | 1 TB | 10 | 32 | $229.00/mo | 2Available | Rent |
FAQ
Questions before you book.
How much memory can I rent for a local LLM?
From 32 GB up to 512 GB of unified memory, shared by the CPU and the GPU.
Can I run MLX and llama.cpp?
Yes. You have admin rights and install them yourself, for example with pip for MLX or Homebrew for llama.cpp.
Will a 70B model fit?
At 4-bit a 70B model needs about 42 GB for weights, and about 20 percent more for context and the runtime. The model fit tool on this page checks it against every config.
Is the machine shared while my model runs?
No. One booking is one physical Mac Studio, so the memory and the GPU are not shared with another customer.
What happens to my model weights when the term ends?
The disk is erased before the Mac is listed again. Keep your own copy of anything you need.
Ready when the work is.
No rack to buy. No machine aging on a desk between releases. Book the config, then stop.