The best local LLM for a Mac Studio M4 Ultra (96 GB) is Qwen 4.1 32B-A3B at 113 tok/s. With 96 GB of unified memory it runs 58 of the models we benchmark — from compact options up to 141B-class models. For everyday chat and coding, Qwen 4.1 32B-A3B is the sweet spot. Full ranking below.
Ranked by LLMCheck suitability (capability balanced against real speed on the M4 Ultra). Click a model for its full benchmark and setup. Speeds marked est. are scaled from measured runs by memory bandwidth.
| # | Model | Size | License | Speed | Capability |
|---|---|---|---|---|---|
| 1 | Qwen 4.1 32B-A3B | 32B | Apache 2.0 | 113 tok/s est. | 46/50 |
| 2 | Qwen 4 | 32B | Apache 2.0 | 109 tok/s est. | 45/50 |
| 3 | Qwen 4 Coder | 32B | Apache 2.0 | 106 tok/s est. | 44/50 |
| 4 | Qwen 4 Preview 32B-A3B | 32B | Apache 2.0 | 106 tok/s est. | 42/50 |
| 5 | Qwen 3.6-35B-A3B | 35B | Apache 2.0 | 95 tok/s est. | 38/50 |
| 6 | Gemma 4 26B-A4B | 26B | Apache 2.0 | 87 tok/s est. | 35/50 |
| 7 | Gemma 4 31B | 31B | Apache 2.0 | 44 tok/s est. | 40/50 |
| 8 | GLM 5.2 Air | 106B | MIT | 38 tok/s | 40/50 |
| 9 | Mistral Medium 4 | 41B | Apache 2.0 | 87 tok/s est. | 34/50 |
| 10 | Phi-5 Large 28B | 28B | MIT | 69 tok/s est. | 36/50 |
| 11 | Gemma 4.5 12B | 12B | Gemma | 136 tok/s est. | 28/50 |
| 12 | Phi-5 Medium 14B | 14B | MIT | 118 tok/s est. | 28/50 |
Showing the top 12 of 58 models that fit in 96 GB. See the full leaderboard or all benchmarks.
The fastest way to get started is Ollama. Install it, then pull the top pick for your Mac:
Prefer a GUI? LM Studio gives you a one-click download and chat window. For step-by-step help see our Ollama install guide, or open the Qwen 4.1 32B-A3B on M4 Ultra benchmark page for exact settings.
The Mac Studio M4 Ultra (96 GB) comfortably runs 58 of the models we benchmark, led by Qwen 4.1 32B-A3B. Grab one and start running LLMs offline today:
As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence our rankings.
Qwen 4.1 32B-A3B (32B, Apache 2.0) is the best all-round pick at 113 tok/s on the M4 Ultra. If you want maximum speed, Gemma 4 E2B hits 282 tok/s; for maximum capability, Qwen 4 still fits in 96 GB.
About 58 of the 79 models in the LLMCheck leaderboard fit in 96 GB of unified memory, from compact models up to Mixtral 8x22B (141B).
Yes. A 70B model in Q4 quantization needs roughly 40–44 GB of memory, which fits in 96 GB with headroom for context.
96 GB is plenty for local AI — you can run capable 30B–70B-class models. Because Apple Silicon uses unified memory, that figure is both your system RAM and your VRAM.