Can I Run Devstral on MacBook Pro 16" (M5 Max, 128 GB)?

Written by Jakub Rusinowski · Last updated August 15, 2026

Yes, but it is tight

Yes, but it is tight — Devstral-2 123B at Q5_K_M needs about 90.9 GB of the 96 GB usable on MacBook Pro 16" (M5 Max, 128 GB), leaving only ~5.1 GB before the runtime starts swapping. Expect ~10.9 tok/s (estimated), with room for about 16,384 tokens of context.

Confidence: low · Recommended quantization: Q5_K_M · Estimated speed: ~10.9 tok/s

MacBook Pro 16" (M5 Max, 128 GB) — what it gives a model

Usable memory for models	96 GB
Memory bandwidth	614 GB/s
Form factor	Laptop
Operating system	macOS
Memory upgradeable	No — soldered

Devstral on MacBook Pro 16" (M5 Max, 128 GB): memory by quantization

Quant	Memory needed	Fits 96 GB?	Max context	Est. speed	Download
F16	249.7 GB	✗ No	—	—	246 GB
Q8_0	134.4 GB	✗ No	—	—	130.7 GB
Q6_K	104.5 GB	✗ No	—	—	100.9 GB
Q5_K_M	90.9 GB	✓ Yes	16K	~10.9 tok/s	87.2 GB
Q4_K_M	77.9 GB	✓ Yes	32K	~12.7 tok/s	74.3 GB
Q3_K_M	56.1 GB	✓ Yes	64K	~17.3 tok/s	52.4 GB
Q2_K	44.1 GB	✓ Yes	64K	~21.6 tok/s	40.4 GB

Which Devstral sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
Devstral-2 123B	77.9 GB	✓ Fits	~12.7 tok/s
Devstral Small 24B	16.6 GB	✓ Fits	~20.8 tok/s
Devstral-2 22B	15.7 GB	✓ Fits	~22.3 tok/s

What to watch out for

This model does not publish its full attention configuration, so the KV-cache share of these figures is inferred from its parameter count rather than computed exactly.
Memory on this machine is not upgradeable, so the configuration you buy is the ceiling for every model you will ever run on it.

MacBook Pro M5 Max 128 GB limitations

Unified memory is soldered and cannot be upgraded after purchase.

Recommended setup

Ollama or LM Studio (Metal) — MLX for the fastest Apple-native throughput

How these numbers are calculated

Memory need = quantized weights + KV cache at 8,192 tokens (f16) + 0.8 GB runtime overhead.
128 GB unified memory at 614 GB/s, shared between CPU and GPU.
macOS reserves a share of unified memory for the system, so not all of it is available to a model.
Throughput is a memory-bandwidth roofline estimate, not a measurement. It is labelled "estimated" everywhere it appears.
KV cache is inferred from the parameter count because this model does not publish its full attention configuration.

FAQ