CanItRun Logocanitrun.

Apple M5 Max

The Apple M5 Max ships in 36–128 GB unified-memory configurations at 460–614 GB/s. Across those configurations it runs 64 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:128 GB64 GB48 GB36 GB
ConfigurationBandwidthCPU coresNative models+ Offload
128 GB614 GB/s18 (6S + 12P)64 / 840
64 GB614 GB/s18 (6S + 12P)56 / 840
48 GB614 GB/s18 (6S + 12P)51 / 840
36 GB460 GB/s18 (6S + 12P)47 / 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 M5 Max'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.

Apple's Max-tier bandwidth was flat for two straight generations

The M5 Max isn't the first bandwidth bump in Apple's laptop lineup, but it lands after a longer stall than the raw "614 GB/s" number suggests on its own. Here's the top memory bandwidth for each generation's fastest Max-tier configuration, back to the M2 Max:

Apple M2 Max (96GB)
400 GB/s
Apple M3 Max (128GB)
400 GB/s
Apple M4 Max (128GB)
546 GB/s
Apple M5 Max (128GB) — this page
614 GB/s

M2 Max and M3 Max shipped the exact same 400 GB/s ceiling two years apart — real architecture changes, zero bandwidth gain. M4 Max broke that streak with a 36.5% jump to 546 GB/s, and M5 Max keeps the momentum going with another 12.5% to 614 GB/s — a 53.5% gain over the M2/M3 generation in total. Since LLM decode is bandwidth-bound, that compounds directly into tokens/sec: the same model at the same quantization runs meaningfully faster on this chip than on an M3 Max of the same 128GB capacity, and the gap is bandwidth, not clever software.

Apple M5 Max (128GB)

With 128 GB LPDDR5X at 614 GB/s, this configuration runs 64 models natively. It runs 70B-class dense models and large MoE models entirely in VRAM.

Apple M5 Max (128GB): the maximum-memory M5 Max configuration, announced March 3, 2026 for the 14-inch and 16-inch MacBook Pro, on the full, un-binned 18-core CPU (6 super cores + 12 performance cores)/40-core GPU die at 614 GB/s — the fastest memory bandwidth Apple has ever shipped in a laptop, 12.5% above the M4 Max's 546 GB/s ceiling (see this page's bandwidth note for the full multi-generation picture).

With 120 GB of real headroom, this is the first M5 Max configuration to fit genuine frontier-scale MoE models: Qwen3 235B-A22B fits at Q2_K (102.05 GB, 16.6 tok/s) and GPT-OSS 120B reaches Q6_K (107.93 GB, 34.4 tok/s) — both roughly 12.5% faster than the identically-sized M4 Max 128GB manages on the same fits (14.8 and 30.6 tok/s there), tracking the 614-vs-546 GB/s bandwidth gap almost exactly. Dense models do well too: Llama 3.3 70B and Qwen2.5 72B both reach Q8_0 (86.35 GB and 88.73 GB), near-full precision, at 6.4 and 6.2 tok/s. 64 of the 84 tracked models fit natively.

MLX and llama.cpp's Metal backend are both fully optimized here, and Apple Silicon's efficiency advantage over a discrete-GPU workstation is largest at this end of the lineup — this much memory bandwidth in a laptop chassis still has no real equivalent on the discrete-GPU side.

Models the 128 GB configuration runs natively (64)

Show 59 more

Apple M5 Max (64GB)

With 64 GB LPDDR5X at 614 GB/s, this configuration runs 56 models natively. It handles 70B-class models at Q4 quantization.

Apple M5 Max (64GB): the mid-tier M5 Max configuration, available in the 14-inch and 16-inch MacBook Pro on the same full 18-core CPU/40-core GPU die as the 48GB and 128GB builds, all sharing the generation's 614 GB/s ceiling — only the 36GB base configuration ships on a slower, cut-down 32-core-GPU die (see this page's bandwidth note).

With 56 GB of real headroom, this is the first M5 Max tier where 70B-class dense models clear a real quantization instead of the aggressive Q2_K floor: Llama 3.3 70B reaches Q4_K_M (50.75 GB, 10.8 tok/s) and Qwen2.5 72B reaches Q4_K_M (52.12 GB, 10.6 tok/s) — both about 12.5% faster than the identically-sized M4 Max 64GB (9.6 and 9.4 tok/s there). Command-R 35B goes further, hitting near-full-precision Q8_0 (53.70 GB, 10.2 tok/s). MoE models are faster still at this size: Qwen 3.5 35B-A3B fits at Q8_0 (41.86 GB, 45.5 tok/s). 56 of the 84 tracked models fit natively.

MLX and llama.cpp's Metal backend are both mature here. Choosing this over the identically-priced-per-GB M5 Pro 64GB buys exactly 2x decode speed on every model both fit (614 vs 307 GB/s) — the Max tier's advantage at this size is bandwidth, not capacity, the same story as the 48GB tier one step down.

Models the 64 GB configuration runs natively (56)

Show 51 more

Apple M5 Max (48GB)

With 48 GB LPDDR5X at 614 GB/s, this configuration runs 51 models natively. It handles 70B-class models at Q4 quantization.

Apple M5 Max (48GB): the entry point into M5 Max's full, un-binned 18-core CPU/40-core GPU die, available in the 14-inch and 16-inch MacBook Pro — the same 614 GB/s as the 64GB and 128GB builds, a real step up from the cut-down 36GB base configuration's 460 GB/s.

With 40 GB of real headroom — identical to the M5 Pro 48GB configuration — this tier fits the same 51 of 84 tracked models, but at exactly double the speed on every shared fit: Llama 3.3 70B reaches the same Q2_K (32.88 GB) at 16.7 tok/s here versus 8.4 tok/s on the M5 Pro 48GB, and Qwen2.5 72B reaches 16.3 tok/s versus 8.2 tok/s — both track the 614-vs-307 GB/s bandwidth ratio exactly. MoE models do particularly well: Qwen 3.5 35B-A3B fits at near-full Q6_K (32.37 GB, 58.6 tok/s), about 12.5% faster than the identically-sized M4 Max 48GB's 52.1 tok/s.

MLX and llama.cpp's Metal backend are both fully supported. Choosing this over the identically-priced-per-GB M5 Pro 48GB buys roughly 2x decode speed on every model both fit — the Max tier's advantage here is bandwidth, not capacity.

Models the 48 GB configuration runs natively (51)

Show 46 more

Apple M5 Max (36GB)

With 36 GB LPDDR5X at 460 GB/s, this configuration runs 47 models natively. It comfortably runs 7B–32B models at Q4; 70B-class models typically need CPU offload.

Apple M5 Max (36GB): the standard base configuration of Apple's M5 Max chip, announced March 3, 2026 for the 14-inch and 16-inch MacBook Pro, on a cut-down 32-core-GPU die at 460 GB/s — 154 GB/s less than the 614 GB/s the 48GB, 64GB, and 128GB configurations get on the full 40-core-GPU die. Apple offers no way to pair 36GB with the faster die, the same binned-base-tier pattern the M3 Max and M4 Max shipped in the two generations before it (see this page's bandwidth note). 36GB is exclusively an M5 Max configuration — Apple doesn't offer it on the M5 Pro chip at all.

With 28 GB of real headroom, this configuration fits the same 47 of 84 tracked models as the identically-sized M3 Max 36GB, M3 Pro 36GB, and M4 Max 36GB — capacity decides what fits, not bandwidth — but runs about 12% faster than the M4 Max 36GB on every shared fit, since 460 GB/s is 12.2% more than that chip's 410 GB/s. Qwen3 32B reaches Q5_K_M (27.66 GB) at 14.9 tok/s here versus 13.3 tok/s on the M4 Max 36GB, and Qwen 3.5 35B-A3B reaches Q4_K_M (24.06 GB) at 58.8 tok/s versus 52.4 tok/s — both track the 460-vs-410 GB/s ratio closely.

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, the same fix and figure this site notes for the identically-sized M3 Max, M3 Pro, and M4 Max 36GB configurations.

Models the 36 GB configuration runs natively (47)

Show 42 more

Too large for any Apple M5 Max configuration (20)

Continue reading

hardware11 min
Apple Silicon for LLMs: M1-M5 Complete Guide

Frequently asked questions

How much memory does the Apple M5 Max have?
The Apple M5 Max ships in 4 unified-memory configurations: 128 GB, 64 GB, 48 GB, 36 GB, at 460–614 GB/s depending on configuration.
Should I get the 36 GB or 128 GB Apple M5 Max?
Both run everything that fits natively in 36 GB. The extra memory in the 128 GB configuration additionally fits Qwen3 235B-A22B (MoE), MiniMax M2.5 229B, MiniMax M2.7 229B, and 14 more models natively in VRAM — worth the upgrade if you plan to run any of those.
How much VRAM does the Apple M5 Max (128GB) have?
The Apple M5 Max (128GB) has 128 GB of LPDDR5X with 614 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M5 Max (128GB) best for?
With 128 GB of unified memory, the Apple M5 Max (128GB) is a high-capacity laptop platform that runs 70B-class dense models and large MoE models natively, with plenty of room for long context.
What LLMs can the Apple M5 Max (128GB) run locally?
The Apple M5 Max (128GB) can run 64 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q8_0, Llama 3.1 8B Instruct at FP32, Llama 3.2 3B Instruct at FP32.
Can the Apple M5 Max (128GB) run Llama 3.3 70B Instruct?
Yes. The Apple M5 Max (128GB) runs Llama 3.3 70B Instruct natively in VRAM at Q8_0 quantization, achieving approximately 6.4 tokens per second.
Show 20 more questions
Can the Apple M5 Max (128GB) run Qwen 3.6 27B?
Yes. The Apple M5 Max (128GB) runs Qwen 3.6 27B natively in VRAM at BF16 quantization, achieving approximately 9 tokens per second.
Can the Apple M5 Max (128GB) run Llama 3.1 8B Instruct?
Yes. The Apple M5 Max (128GB) runs Llama 3.1 8B Instruct natively in VRAM at FP32 quantization, achieving approximately 14.9 tokens per second.
How much VRAM does the Apple M5 Max (64GB) have?
The Apple M5 Max (64GB) has 64 GB of LPDDR5X with 614 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M5 Max (64GB) best for?
With 64 GB of VRAM, the Apple M5 Max (64GB) is ideal for running 70B-class models at Q4 quantization and large MoE models — a workstation sweet spot for local inference.
What LLMs can the Apple M5 Max (64GB) run locally?
The Apple M5 Max (64GB) can run 56 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q4_K_M, Llama 3.1 8B Instruct at FP32, Llama 3.2 3B Instruct at FP32.
Can the Apple M5 Max (64GB) run Llama 3.3 70B Instruct?
Yes. The Apple M5 Max (64GB) runs Llama 3.3 70B Instruct natively in VRAM at Q4_K_M quantization, achieving approximately 10.8 tokens per second.
Can the Apple M5 Max (64GB) run Qwen 3.6 27B?
Yes. The Apple M5 Max (64GB) runs Qwen 3.6 27B natively in VRAM at Q8_0 quantization, achieving approximately 16.8 tokens per second.
Can the Apple M5 Max (64GB) run Llama 3.1 8B Instruct?
Yes. The Apple M5 Max (64GB) runs Llama 3.1 8B Instruct natively in VRAM at FP32 quantization, achieving approximately 14.9 tokens per second.
How much VRAM does the Apple M5 Max (48GB) have?
The Apple M5 Max (48GB) has 48 GB of LPDDR5X with 614 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M5 Max (48GB) best for?
With 48 GB of VRAM, the Apple M5 Max (48GB) is ideal for running 70B-class models at Q4 quantization and large MoE models — a workstation sweet spot for local inference.
What LLMs can the Apple M5 Max (48GB) run locally?
The Apple M5 Max (48GB) can run 51 of the 84 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q2_K, Llama 3.1 8B Instruct at FP32, Llama 3.2 3B Instruct at FP32.
Can the Apple M5 Max (48GB) run Llama 3.3 70B Instruct?
Yes. The Apple M5 Max (48GB) runs Llama 3.3 70B Instruct natively in VRAM at Q2_K quantization, achieving approximately 16.7 tokens per second.
Can the Apple M5 Max (48GB) run Qwen 3.6 27B?
Yes. The Apple M5 Max (48GB) runs Qwen 3.6 27B natively in VRAM at Q8_0 quantization, achieving approximately 16.8 tokens per second.
Can the Apple M5 Max (48GB) run Llama 3.1 8B Instruct?
Yes. The Apple M5 Max (48GB) runs Llama 3.1 8B Instruct natively in VRAM at FP32 quantization, achieving approximately 14.9 tokens per second.
How much VRAM does the Apple M5 Max (36GB) have?
The Apple M5 Max (36GB) has 36 GB of LPDDR5X with 460 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
What is the Apple M5 Max (36GB) best for?
With 36 GB of VRAM, the Apple M5 Max (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 M5 Max (36GB) run locally?
The Apple M5 Max (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 M5 Max (36GB) run Llama 3.3 70B Instruct?
The Apple M5 Max (36GB) 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 M5 Max (36GB) run Qwen 3.6 27B?
Yes. The Apple M5 Max (36GB) runs Qwen 3.6 27B natively in VRAM at Q6_K quantization, achieving approximately 16.2 tokens per second.
Can the Apple M5 Max (36GB) run Llama 3.1 8B Instruct?
Yes. The Apple M5 Max (36GB) runs Llama 3.1 8B Instruct natively in VRAM at BF16 quantization, achieving approximately 21.6 tokens per second.