CanItRun Logocanitrun.

Apple M3 Pro

The Apple M3 Pro ships in 18–36 GB unified-memory configurations at 150 GB/s. Across those configurations it runs 47 of our 84 tracked models natively in VRAM at 8k context.

More memory means more of our tracked models fit natively — see which configuration you need below.

Jump to:36 GB18 GB
ConfigurationBandwidthCPU coresNative models+ Offload
36 GB150 GB/s12 (6P + 6E)47 / 840
18 GB150 GB/s11 (5P + 6E)28 / 840
Vendor: Apple
Memory type: LPDDR5
Compute backend: METAL
Software: MLX gives the best performance on Apple Silicon; llama.cpp Metal backend is a solid alternative. Both are well-supported by Ollama.

How much of the Apple M3 Pro's memory is actually usable?

macOS and background apps need a slice of the pool before a model gets to use it — this site reserves 8GB on every unified-memory GPU, the same baseline used everywhere else on this site. What's left is real headroom for a model's weights and KV cache:

macOS + background apps (8 GB reserved)usable for model weights + KV cacheVertical lines mark roughly where a 7B/14B/32B/... dense model at Q4_K_M lands.

The one Apple Silicon tier that got slower before it got faster

M3 Pro is the only tier in the whole Apple Silicon lineup where a new generation shipped with less memory bandwidth than the one it replaced — Apple narrowed the memory bus from 256-bit to 192-bit, and the cut went unannounced until reviewers benchmarked it. Tracking the Pro tier's bandwidth across five generations shows the dip, and the recovery that followed it:

Apple M1 Pro (32GB)
200 GB/s
Apple M2 Pro (32GB)
200 GB/s
Apple M3 Pro (36GB) — this page
150 GB/s
Apple M4 Pro (48GB)
273 GB/s
Apple M5 Pro (48GB)
307 GB/s

M1 Pro and M2 Pro both shipped 200 GB/s — no change across a full generation. M3 Pro then dropped 25% to 150 GB/s, a real regression MacRumors and others confirmed within days of launch. M4 Pro reversed it hard, jumping 82% to 273 GB/s, and M5 Pro added another 12.5% to reach 307 GB/s. Net result: M5 Pro ends up 53.5% ahead of where M2 Pro left off — the same overall gain the Max tier posted over the same stretch (M2/M3 Max's 400 GB/s to M5 Max's 614 GB/s) despite the Pro tier's detour through a real step backward in the middle.

Apple M3 Pro (36GB)

With 36 GB LPDDR5 at 150 GB/s, this configuration runs 47 models natively. It comfortably runs 7B–32B models at Q4; 70B-class models typically need CPU offload.

Apple M3 Pro (36GB): the 36GB M3 Pro configuration, standard on the 12-core CPU (6P+6E)/18-core GPU upgrade die, at the same 150 GB/s bandwidth as the 18GB build — a real 25% cut from the M1 Pro/M2 Pro's 200 GB/s (see this page's bandwidth note for the full multi-generation picture).

With 28 GB of real headroom, this configuration fits the same 47 of the 84 models as the identically-sized 36GB M3 Max — capacity decides what fits, not bandwidth — but at half the speed on every model both share, since 150 GB/s is half the M3 Max's 300 GB/s binned-die bandwidth. Qwen 3.5 35B-A3B reaches Q4_K_M (24.06 GB) at 19.2 tok/s here versus 38.4 tok/s on the 36GB M3 Max, and Qwen3 32B reaches Q5_K_M (27.66 GB) at 4.9 tok/s versus 9.7 tok/s — both almost exactly half.

MLX and llama.cpp's Metal backend are both mature here. The default macOS Metal working-set limit (~75% of 36 GB, 27 GB) sits just under this page's largest fit — Qwen3 32B's 27.66 GB Q5_K_M build — so raising it with `sudo sysctl iogpu.wired_limit_mb=28672` (leaving 8 GB for macOS) covers it with room to spare.

Models the 36 GB configuration runs natively (47)

Show 42 more

Apple M3 Pro (18GB)

With 18 GB LPDDR5 at 150 GB/s, this configuration runs 28 models natively. It handles smaller models (7B–14B) at Q4–Q5 quantization.

Apple M3 Pro (18GB): the base 14-inch MacBook Pro's M3 Pro configuration, launched November 2023 with the 11-core CPU (5P+6E)/14-core GPU die — the 12-core/18-core upgrade ships standard with the 36GB configuration instead. Both dies share the same 150 GB/s bandwidth, a real 25% cut from the 200 GB/s the M1 Pro and M2 Pro both shipped at (see this page's bandwidth note) — Apple narrowed the memory bus from 256-bit to 192-bit and never explained why.

With 10 GB of real headroom — 2 GB more than the M1 Pro/M2 Pro 16GB tier's 8 GB, since Apple's 3-memory-channel M3 Pro design starts at 18GB rather than 16GB — this configuration handles 12-14B models at real quality rather than the bottom of the quant ladder: Qwen3 14B fits at Q3_K_M (9.48 GB, 14.2 tok/s) and Gemma 3 12B fits at Q4_K_M (9.51 GB, 14.1 tok/s). Llama 3.1 8B and Qwen2.5 7B both clear Q6_K (8.56 GB and 8.71 GB, roughly 15.5 tok/s). 28 of the 84 tracked models fit natively.

MLX and llama.cpp's Metal backend are both mature here. Nothing this configuration fits comes close to the default ~13.5 GB Metal working-set limit (75% of 18 GB) — the largest fit is under 10 GB — so the manual wired_limit override doesn't apply at this capacity.

Models the 18 GB configuration runs natively (28)

Show 23 more

Too large for any Apple M3 Pro configuration (37)

Compare Apple M3 Pro with other GPUs

Explore

Browse all guides →Browse all models →Browse all GPUs →

Frequently asked questions

How much memory does the Apple M3 Pro have?
The Apple M3 Pro ships in 2 unified-memory configurations: 36 GB and 18 GB, all at 150 GB/s.
Should I get the 18 GB or 36 GB Apple M3 Pro?
Both run everything that fits natively in 18 GB. The extra memory in the 36 GB configuration additionally fits Mixtral 8x7B Instruct v0.1, Command-R 35B, Qwen 3.5 35B-A3B (MoE), and 16 more models natively in VRAM — worth the upgrade if you plan to run any of those.
How much VRAM does the Apple M3 Pro (36GB) have?
The Apple M3 Pro (36GB) has 36 GB of LPDDR5 with 150 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M3 Pro (36GB) best for?
With 36 GB of VRAM, the Apple M3 Pro (36GB) is well-suited for running 7B–32B models at Q4 with room for context, making it a great all-rounder for local LLM inference.
What LLMs can the Apple M3 Pro (36GB) run locally?
The Apple M3 Pro (36GB) can run 47 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.1 8B Instruct at BF16, Llama 3.2 3B Instruct at FP32, Llama 3.2 1B Instruct at FP32.
Can the Apple M3 Pro (36GB) run Llama 3.3 70B Instruct?
The Apple M3 Pro (36GB) does not have enough VRAM to run Llama 3.3 70B Instruct. You would need more VRAM or a lower quantization level.
Show 8 more questions
Can the Apple M3 Pro (36GB) run Qwen 3.6 27B?
Yes. The Apple M3 Pro (36GB) runs Qwen 3.6 27B natively in VRAM at Q6_K quantization, achieving approximately 5.3 tokens per second.
Can the Apple M3 Pro (36GB) run Llama 3.1 8B Instruct?
Yes. The Apple M3 Pro (36GB) runs Llama 3.1 8B Instruct natively in VRAM at BF16 quantization, achieving approximately 7 tokens per second.
How much VRAM does the Apple M3 Pro (18GB) have?
The Apple M3 Pro (18GB) has 18 GB of LPDDR5 with 150 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M3 Pro (18GB) best for?
With 18 GB of VRAM, the Apple M3 Pro (18GB) handles smaller models (7B–14B) at Q4–Q5 quantization — ideal for entry-level local LLM experimentation and lightweight inference.
What LLMs can the Apple M3 Pro (18GB) run locally?
The Apple M3 Pro (18GB) can run 28 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.1 8B Instruct at Q6_K, Llama 3.2 3B Instruct at BF16, Llama 3.2 1B Instruct at FP32.
Can the Apple M3 Pro (18GB) run Llama 3.3 70B Instruct?
The Apple M3 Pro (18GB) does not have enough VRAM to run Llama 3.3 70B Instruct. You would need more VRAM or a lower quantization level.
Can the Apple M3 Pro (18GB) run Qwen 3.6 27B?
The Apple M3 Pro (18GB) does not have enough VRAM to run Qwen 3.6 27B. You would need more VRAM or a lower quantization level.
Can the Apple M3 Pro (18GB) run Llama 3.1 8B Instruct?
Yes. The Apple M3 Pro (18GB) runs Llama 3.1 8B Instruct natively in VRAM at Q6_K quantization, achieving approximately 15.7 tokens per second.