Can I Run InternLM 3 on MacBook Pro 14" (M4 Pro, 24 GB)?

Written by Jakub Rusinowski · Last updated January 15, 2025

Yes, but it is tight

Yes, but it is tight — InternLM 3 20B Instruct at Q5_K_M needs about 16.6 GB of the 18 GB usable on MacBook Pro 14" (M4 Pro, 24 GB), leaving only ~1.4 GB before the runtime starts swapping. Expect ~9.7 tok/s (estimated), with room for about 8,192 tokens of context.

Confidence: low · Recommended quantization: Q5_K_M · Estimated speed: ~9.7 tok/s

MacBook Pro 14" (M4 Pro, 24 GB) — what it gives a model

Usable memory for models	18 GB
Memory bandwidth	273 GB/s
Form factor	Laptop
Operating system	macOS
Memory upgradeable	No — soldered
Price	$1,999 (lib/data/laptops.ts (street price), checked 2026-07-06)

InternLM 3 on MacBook Pro 14" (M4 Pro, 24 GB): memory by quantization

Quant	Memory needed	Fits 18 GB?	Max context	Est. speed	Download
F16	42.4 GB	✗ No	—	—	40 GB
Q8_0	23.6 GB	✗ No	—	—	21.3 GB
Q6_K	18.8 GB	✗ No	—	—	16.4 GB
Q5_K_M	16.6 GB	✓ Yes	8K	~9.7 tok/s	14.2 GB
Q4_K_M	14.5 GB	✓ Yes	16K	~11.2 tok/s	12.1 GB
Q3_K_M	10.9 GB	✓ Yes	32K	~15.3 tok/s	8.5 GB
Q2_K	9 GB	✓ Yes	32K	~19.2 tok/s	6.6 GB

Which InternLM 3 sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
InternLM 3 20B Instruct	14.5 GB	✓ Fits	~11.2 tok/s
InternLM 3 8B Instruct	6.5 GB	✓ Fits	~25.1 tok/s

What to watch out for

Only ~1.4 GB of headroom at Q5_K_M: a longer context or a second application can push this into swapping.
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 M4 Pro 24 GB limitations

Unified memory is soldered and cannot be upgraded after purchase.
24 GB is the practical ceiling for a single mid-size model plus a long context.

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.
24 GB unified memory at 273 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