CanItRun Logocanitrun.

Apple M4

The Apple M4 ships in 16–32 GB unified-memory configurations at 120 GB/s. Across those configurations it runs 46 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:32 GB16 GB
ConfigurationBandwidthCPU coresNative models+ Offload
32 GB120 GB/s10 (4P + 6E)46 / 840
16 GB120 GB/s10 (4P + 6E)26 / 840
Vendor: Apple
Memory type: LPDDR5X
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 M4'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.

Base-tier bandwidth: a steady climb, with one flat generation

Apple's cheapest M-series chip doesn't get a bandwidth bump every year. Tracking the base tier's memory bandwidth across five generations:

Apple M1 (16GB)
68 GB/s
Apple M2 (16GB)
100 GB/s
Apple M3 (16GB)
100 GB/s
Apple M4 (16GB) — this page
120 GB/s
Apple M5 (16GB)
153 GB/s

M1's 68 GB/s jumped 47% to M2's 100 GB/s, but M3 (this generation's immediate predecessor) shipped at the same 100 GB/s — the only base-tier generation that didn't move bandwidth at all. M4 (this page) resumed the climb, up 20% to 120 GB/s, and M5 added another 27.5% to reach 153 GB/s. Since decode is bandwidth-bound, an M4 (this page) decodes the same model about 20% faster than an identically-sized M3, at identical capacity — worth knowing since "base M3" and "base M4" otherwise look like a minor spec bump on paper.

Apple M4 (32GB)

With 32 GB LPDDR5X at 120 GB/s, this configuration runs 46 models natively. It comfortably runs 7B–32B models at Q4; 70B-class models typically need CPU offload.

Apple M4 (32GB): the maximum-memory configuration of Apple's entry-level M4 chip, launched October 2024 and configurable in the MacBook Pro and MacBook Air, with a 10-core CPU (4P+6E) and 120 GB/s of unified-memory bandwidth — a 20% jump over the M2 and M3 base chips' identical 100 GB/s (the M3 generation's one tier that didn't move bandwidth at all from its predecessor).

With 24 GB of real headroom, this configuration is the first base-tier M4 build to reach genuine mid-size models: Llama 3.1 8B and Qwen2.5 7B both fit at full BF16 precision (19.12 GB and 17.55 GB) at 5.6 and 6.1 tok/s, and Qwen3 32B fits at Q4_K_M (23.88 GB, 4.5 tok/s). The MoE-vs-dense gap is stark at this capacity: Qwen 3.5 35B-A3B — a bigger model by parameter count — fits at Q3_K_M (19.04 GB) at 19.3 tok/s, more than 4x Qwen3 32B's speed, since MoE decode only reads its active parameters off the memory bus. 46 of the 84 tracked models fit natively.

MLX and llama.cpp's Metal backend are both mature on this chip. This configuration's largest fits (~20-24 GB) stay under the default ~24 GB Metal working-set limit (75% of 32 GB) in most cases, so the manual wired_limit override rarely matters here.

Models the 32 GB configuration runs natively (46)

Show 41 more

Apple M4 (16GB)

With 16 GB LPDDR5X at 120 GB/s, this configuration runs 26 models natively. It handles smaller models (7B–14B) at Q4–Q5 quantization.

Apple M4 (16GB): the base configuration of Apple's entry-level M4 chip, launched October 2024 as the standard memory across the MacBook Air, MacBook Pro, Mac mini, and iMac, with the same 10-core CPU (4P+6E) and 120 GB/s bandwidth as the 32GB build — only the memory pool changes.

With 8 GB of real headroom after the standard 8 GB reservation, this site's calculator fits Llama 3.1 8B at Q5_K_M (7.58 GB, 14.2 tok/s) and Qwen2.5 7B at Q6_K (7.51 GB, 14.3 tok/s) — about 20% faster than the identically-sized M3 16GB on the same two models (11.8 and 11.9 tok/s there), tracking the 120-vs-100 GB/s bandwidth ratio closely. 26 of the 84 tracked models fit natively — the same count as the base M1, M2, and M3 16GB configurations at this identical 8 GB headroom, since capacity, not bandwidth, decides the model list.

MLX and llama.cpp's Metal backend are both mature on this chip. Nothing this configuration fits comes close to the default ~12 GB Metal working-set limit (75% of 16 GB) — the largest fit here is under 8 GB — so the manual wired_limit override other Apple Silicon pages need doesn't apply at this capacity.

Models the 16 GB configuration runs natively (26)

Show 21 more

Too large for any Apple M4 configuration (38)

Explore

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

Frequently asked questions

How much memory does the Apple M4 have?
The Apple M4 ships in 2 unified-memory configurations: 32 GB and 16 GB, all at 120 GB/s.
Should I get the 16 GB or 32 GB Apple M4?
Both run everything that fits natively in 16 GB. The extra memory in the 32 GB configuration additionally fits Mixtral 8x7B Instruct v0.1, Qwen 3.5 35B-A3B (MoE), Qwen 3.6 35B, and 17 more models natively in VRAM — worth the upgrade if you plan to run any of those.
How much VRAM does the Apple M4 (32GB) have?
The Apple M4 (32GB) has 32 GB of LPDDR5X with 120 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M4 (32GB) best for?
With 32 GB of VRAM, the Apple M4 (32GB) 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 M4 (32GB) run locally?
The Apple M4 (32GB) can run 46 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 M4 (32GB) run Llama 3.3 70B Instruct?
The Apple M4 (32GB) 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 M4 (32GB) run Qwen 3.6 27B?
Yes. The Apple M4 (32GB) runs Qwen 3.6 27B natively in VRAM at Q5_K_M quantization, achieving approximately 4.9 tokens per second.
Can the Apple M4 (32GB) run Llama 3.1 8B Instruct?
Yes. The Apple M4 (32GB) runs Llama 3.1 8B Instruct natively in VRAM at BF16 quantization, achieving approximately 5.6 tokens per second.
How much VRAM does the Apple M4 (16GB) have?
The Apple M4 (16GB) has 16 GB of LPDDR5X with 120 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M4 (16GB) best for?
With 16 GB of VRAM, the Apple M4 (16GB) handles smaller models (7B–14B) at Q4–Q5 quantization — ideal for entry-level local LLM experimentation and lightweight inference.
What LLMs can the Apple M4 (16GB) run locally?
The Apple M4 (16GB) can run 26 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.1 8B Instruct at Q5_K_M, Llama 3.2 3B Instruct at Q8_0, Llama 3.2 1B Instruct at FP32.
Can the Apple M4 (16GB) run Llama 3.3 70B Instruct?
The Apple M4 (16GB) 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 M4 (16GB) run Qwen 3.6 27B?
The Apple M4 (16GB) does not have enough VRAM to run Qwen 3.6 27B. You would need more VRAM or a lower quantization level.
Can the Apple M4 (16GB) run Llama 3.1 8B Instruct?
Yes. The Apple M4 (16GB) runs Llama 3.1 8B Instruct natively in VRAM at Q5_K_M quantization, achieving approximately 14.2 tokens per second.