Is the M4 Pro faster than the M3 Pro for LLMs?

The M4 Pro reads memory about 82% faster than the M3 Pro (273 vs 150 GB/s), so it writes about 82% faster with any model that fits both. Reading a long prompt depends on GPU compute instead, which this comparison doesn’t cover.

Which can run bigger models, the M3 Pro or the M4 Pro?

The M4 Pro: The M4 Pro goes up to 64 GB of memory, the M3 Pro to 36 GB. At its largest it runs 31 of 73 open models well, against 28.

M3 Pro vs M4 Pro for local LLMs: memory, speed, which to buy

Side by side

	M3 Pro	M4 Pro
Memory sizes	18, 36 GB	24, 48, 64 GB
Memory bandwidth	150 GB/s	273 GB/s
For models, at most	about 27 GB	about 48 GB
Models that run well	28 of 73	31 of 73
MLX seat class	Pro	Pro
Found in	MacBook Pro (14-inch, M3 Pro or M3 Max, Nov 2023), MacBook Pro (16-inch, Nov 2023)	Mac mini (2024), MacBook Pro (14-inch, M4 Pro or M4 Max, 2024), MacBook Pro (16-inch, 2024)

Speed with popular models

Tokens a second for one conversation, each chip at its largest memory size (36 GB and 64 GB), with the best version of the model that fits.

Model	M3 Pro	M4 Pro
Llama 3.1 8B Instruct8B	10–18 6-bit	14–26 8-bit
gpt-oss-20b21B · 3.6B active	36–63 native 4.25-bit	65–115 native 4.25-bit
Qwen3.6 35B A3B35B · 3B active	40–71 4-bit	38–68 8-bit
Qwen3.8 27B27B	3–5 6-bit	4–8 8-bit
Llama 3.3 70B Instruct70B	3–5 2-bit lower quality	3–6 4-bit
gpt-oss-120b117B · 5.1B active	too big	77–137 2-bit lower quality
Qwen3.5 122B A10B122B · 10B active	too big	40–70 2-bit lower quality
DeepSeek V4 Flash 0731284B · 13B active	too big	too big

Questions

Is the M4 Pro faster than the M3 Pro for LLMs?: The M4 Pro reads memory about 82% faster than the M3 Pro (273 vs 150 GB/s), so it writes about 82% faster with any model that fits both. Reading a long prompt depends on GPU compute instead, which this comparison doesn’t cover.
Which can run bigger models, the M3 Pro or the M4 Pro?: The M4 Pro: The M4 Pro goes up to 64 GB of memory, the M3 Pro to 36 GB. At its largest it runs 31 of 73 open models well, against 28.

More comparisons

The M3 Pro for local AI · The M4 Pro for local AI · Compare every model · The best Mac for local LLMs

How it’s worked out

Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.