LLM Roofline
Apple · UMA · Backend: mlx

Apple MacBook Pro M3 Max (128 GB) — LLM Sizing & Speed Matrix

128 GB portable memory pool capable of running Llama 4 Scout 109B and Qwen3 235B IQ3/Q4.

Reference Pricing
$4,299 Street
Usable Memory
99.84 GiB (112.64 tuned)
Memory Bandwidth
400 GB/s (76% eff.)
FP16 Compute
28.4 TFLOPS
Roofline Ridge
54.2 FLOP/B
Power (Load / Idle)
115W / 12W

Buy or Rent Apple MacBook Pro M3 Max (128 GB)

Vendor & Cloud Links
MacBook Pro Max on Amazon
Amazon · Direct Hardware
↗
Mac Studio M4 Max at B&H Photo
B&H Photo · Direct Hardware
↗

All 32 Open-Weight Models on Apple MacBook Pro M3 Max (128 GB) (Q4_K_M @ 32k)

Model Status Weights Total @ 32k Decode TTFT (4k) Max Safe Ctx
Mixtral 8x22B Instruct COMFORTABLE 81.2 GiB 89.6 GiB 8.5 tok/s 20.13s 64k
Command A 111B COMFORTABLE 63.9 GiB 68.9 GiB 4.3 tok/s 56.84s 256k
Llama 4 Scout 109B-A17B COMFORTABLE 62.8 GiB 70.0 GiB 15.8 tok/s 8.99s 259k
Qwen2.5 72B Instruct COMFORTABLE 41.9 GiB 53.5 GiB 5.5 tok/s 37.53s 32k
DeepSeek-R1 Distill Llama 70B COMFORTABLE 40.7 GiB 52.3 GiB 5.6 tok/s 36.48s 128k
Llama 3.3 70B Instruct COMFORTABLE 40.7 GiB 52.3 GiB 5.6 tok/s 36.48s 128k
Llama 3.1 70B Instruct COMFORTABLE 40.7 GiB 52.3 GiB 5.6 tok/s 36.48s 128k
Llama-3.3 Nemotron Super 49B COMFORTABLE 28.7 GiB 36.5 GiB 8.1 tok/s 25.67s 128k
Mixtral 8x7B Instruct COMFORTABLE 26.9 GiB 32.0 GiB 21.8 tok/s 6.72s 32k
Qwen3 32B COMFORTABLE 18.9 GiB 28.1 GiB 10.5 tok/s 17.41s 40k
Qwen2.5 Coder 32B COMFORTABLE 18.9 GiB 28.1 GiB 10.5 tok/s 17.02s 32k
DeepSeek-R1 Distill Qwen 32B COMFORTABLE 18.9 GiB 28.1 GiB 10.5 tok/s 17.02s 128k
QwQ 32B Reasoning COMFORTABLE 18.9 GiB 28.1 GiB 10.5 tok/s 17.02s 40k
GLM-4 32B 0414 COMFORTABLE 18.6 GiB 21.9 GiB 13.8 tok/s 16.86s 32k
Qwen3 30B-A3B COMFORTABLE 17.6 GiB 21.4 GiB 50.8 tok/s 2.08s 40k
Gemma 3 27B IT COMFORTABLE 15.8 GiB 20.0 GiB 15.2 tok/s 14.18s 128k
Mistral Small 3.1 24B COMFORTABLE 13.8 GiB 20.1 GiB 15.0 tok/s 12.30s 128k
Codestral 22B COMFORTABLE 12.8 GiB 21.2 GiB 14.3 tok/s 11.78s 32k
Qwen3 14B COMFORTABLE 8.5 GiB 14.8 GiB 20.9 tok/s 7.81s 40k
Qwen2.5 14B Instruct COMFORTABLE 8.5 GiB 15.8 GiB 19.5 tok/s 7.90s 32k
DeepSeek-R1 Distill Qwen 14B COMFORTABLE 8.5 GiB 15.8 GiB 19.5 tok/s 7.90s 128k
Phi-4 14B COMFORTABLE 8.5 GiB 12.8 GiB 24.4 tok/s 7.76s 16k
Gemma 3 12B IT COMFORTABLE 7.0 GiB 10.4 GiB 30.3 tok/s 6.50s 128k
Qwen3 8B COMFORTABLE 4.7 GiB 10.3 GiB 30.7 tok/s 4.41s 40k
Llama 3.1 8B Instruct COMFORTABLE 4.6 GiB 9.7 GiB 32.8 tok/s 4.30s 128k
Qwen2.5 7B Instruct COMFORTABLE 4.4 GiB 7.2 GiB 46.2 tok/s 4.02s 32k
Gemma 3 4B IT COMFORTABLE 2.5 GiB 4.2 GiB 84.9 tok/s 2.31s 128k
DeepSeek-R1 671B OOM 386.4 GiB 390.1 GiB OOM — OOM
DeepSeek-V3 671B OOM 386.4 GiB 390.1 GiB OOM — OOM
Llama 3.1 405B Instruct OOM 233.8 GiB 252.3 GiB OOM — OOM
Llama 4 Maverick 400B-A17B OOM 231.5 GiB 238.8 GiB OOM — OOM
Qwen3 235B-A22B OOM 135.4 GiB 142.4 GiB OOM — OOM

Interactive Apple MacBook Pro M3 Max (128 GB) Simulator

141B (39B active) · GQA
400 GB/s · 28.4 TFLOPS
3. Weight Quantization 4.85 bpw
4. KV Cache
32,768 tok
Compute Roofline
COMFORTABLE FIT Memory-Bandwidth Bound
Max Safe Ctx: 64k
Memory Allocation Footprint 89.60 / 99.84 GiB usable
Weights: 81.20 GiB KV Cache: 7.00 GiB Overhead: 1.39 GiB Headroom: 10.24 GiB
Decode Speed
8.5
tok/s (batch=1)
Prefill Speed
204
prompt tok/s
TTFT Latency
20.13s
for 4,096 tok
Active Params
39.0B
of 141B total
llama.cpp Server Command 
llama-server -hf mistralai/Mixtral-8x22B-Instruct-v0.1:Q4_K_M -ngl 99 -fa on -c 32768 --port 8080
Context KV Cache Total VRAM Decode Status
4k 0.88 GiB 83.05 GiB 10.7 tok/s COMFORTABLE
8k 1.75 GiB 84.35 GiB 10.3 tok/s COMFORTABLE
16k 3.50 GiB 86.10 GiB 9.6 tok/s COMFORTABLE
32k 7.00 GiB 89.60 GiB 8.5 tok/s COMFORTABLE
64k 14.00 GiB 96.60 GiB 6.8 tok/s TIGHT

Buy-vs-Cloud-vs-API Monthly TCO Break-Even

BUY LOCAL
Own Hardware/mo
$89.59
560.7h active + power
Cloud GPU/mo
$2250.21
@$3.49/hr spot/pod
Serverless API/mo
$190.00
Payback: 15.7 mo

Recommended Hardware & Cloud Deployment

Contextual to fit status

Hardware FAQ — Apple MacBook Pro M3 Max (128 GB)

How much usable VRAM / GPU memory does Apple MacBook Pro M3 Max (128 GB) actually provide for LLMs?

Out of the box, Apple MacBook Pro M3 Max (128 GB) safely reserves 99.84 GiB for GPU inference, expandable up to 112.64 GiB with OS UMA tuning.

What is the largest Q4_K_M model Apple MacBook Pro M3 Max (128 GB) can run at 32k context?

With Q4_K_M weights and FP16 KV cache at 32k context, the largest fitting model in our verified catalog on Apple MacBook Pro M3 Max (128 GB) is Mixtral 8x22B Instruct (141B total).