Rig score · mac
Apple M3 Ultra (256GB)
Runs 121 of 135 models (90%)
Elite, runs the biggest open models, including 70B.
Biggest it can run: DeepSeek-V4-Flash
- Grade
- S
- Runnable
- 121
- Total
- 135
- Coverage
- 90%
What Apple M3 Ultra (256GB) can run
At 256 GB of unified memory, Apple M3 Ultra (256GB) runs 121 of 135 tracked local models, 90% coverage that earns a S. The largest it handles is DeepSeek-V4-Flash at 284B parameters; anything bigger needs more memory or a lower quantization.
Apple Silicon shares one 256 GB pool between CPU and GPU, so almost all of it is available to a model. MLX and llama.cpp Metal builds give the best speed here; close memory-heavy apps before loading a large model.
Top models on Apple M3 Ultra (256GB)
FAQ
What local AI rig score does Apple M3 Ultra (256GB) get?
Apple M3 Ultra (256GB) scores S: it runs 121 of 135 tracked local LLMs (90%). The biggest it can run is DeepSeek-V4-Flash.
How big a model can Apple M3 Ultra (256GB) run?
Apple M3 Ultra (256GB) runs models up to DeepSeek-V4-Flash (284B parameters) at Q4_K_M. Its memory is shared with the system, so leave a few GB for the OS.
Is Apple M3 Ultra (256GB) good for running local LLMs?
Apple Silicon shares one 256 GB pool between CPU and GPU, so almost all of it is available to a model. MLX and llama.cpp Metal builds give the best speed here; close memory-heavy apps before loading a large model.
Want your own? Check your rig on the homepage →
Score from coverage of 135 tracked models, estimated with our memory methodology.