What can a Mac Studio M2 Ultra 192GB run?

192GB · 800 GB/s · 150W · $4999

Last-gen flag bearer — a 192 GB workhorse for big unified-memory inference at a more negotiable price.

Runs great32

Comfortable headroom — daily-driver territory.

Model	Quant	Est. speed	Max context
deepseek moe 16b chat	FP16	Infinity tok/s	—
qwen2 7b instruct	FP16	Infinity tok/s	—
Qwen3-0.6B	FP16	399 tok/s	33k
Qwen3.8 Flash Next	Q4_K_M	272 tok/s	141k
Gemma-3n-E4B	FP16	177 tok/s	131k
SmolLM2-1.7B	FP16	146 tok/s	8k
GPT-OSS-120B	Q6_K	118 tok/s	131k
Llama-3.2-3B	FP16	77 tok/s	131k
Qwen3-30B-A3B	FP16	75 tok/s	33k
Qwen3-Coder-30B-A3B	FP16	75 tok/s	262k
GPT-OSS-20B	FP16	69 tok/s	131k
Phi-4-mini	FP16	65 tok/s	131k
Gemma-3-4B	FP16	64 tok/s	131k
Qwen3-4B	FP16	63 tok/s	33k
Mistral-7B-Instruct	FP16	34 tok/s	33k
DeepSeek-R1-Distill-Qwen-7B	FP16	33 tok/s	131k
Qwen2.5-Coder-7B	FP16	33 tok/s	131k
Llama-3.1-8B	FP16	31 tok/s	131k
Qwen3-8B	FP16	31 tok/s	33k
Qwen3-14B	Q8_0	28 tok/s	33k
Codestral-22B	Q8_0	21 tok/s	33k
Mistral-Small-24B	Q8_0	19 tok/s	131k
Gemma-3-12B	FP16	19 tok/s	131k
DeepSeek-R1-Distill-Qwen-32B	Q6_K	18 tok/s	131k
GLM-4-32B	Q6_K	18 tok/s	131k
Qwen3-32B	Q6_K	18 tok/s	33k
Qwen3.8	Q8_0	17 tok/s	262k
Qwen3.8 27B	Q8_0	17 tok/s	262k
Gemma-3-27B	Q8_0	17 tok/s	131k
qwen35 27B	Q8_0	17 tok/s	710k
Phi-4-14B	FP16	17 tok/s	16k
DeepSeek-R1-Distill-Qwen-14B	FP16	17 tok/s	131k

Tight squeeze5

It fits, but close the browser tabs and mind the context.

Model	Quant	Est. speed	Max context
Qwen3-235B-A22B	MLX 4-bit	40 tok/s	22k
Llama-3.3-70B	Q8_0	7 tok/s	131k
Qwen2.5-72B	Q8_0	6 tok/s	131k
Deepseek V4 Flash 0731	IQ2_M	5 tok/s	183k
glm-4.6-air 106B	Q8_0	4 tok/s	125k

Measured on this card0

Everything here was benchmarked on this exact card by someone who owns one — the tables above are our arithmetic; these numbers came off real machines.

Nobody has measured anything on this card yet. Paste a benchmark — it takes one keystroke and no form.

Try your own machine →← All hardware