LocalLLMChecker

Unified memory · Apple

What AI models can an Apple M4 Pro 24GB run?

24 GB is the enthusiast standard, and it shows: a 32B model fits at a good quantization with a long context, and a 70B is reachable if you are willing to drop precision. Memory is shared with the CPU, so the practical ceiling is lower than the sticker number: macOS hands roughly three quarters of it to the GPU by default.

Memory24 GB
Bandwidth273 GB/s
Usable for a model16.8 GB
Runtime backendMETAL

The short answer

Assuming an 8k context window and default settings, these are the models worth downloading first.

Best all-rounder

GPT OSS 20B

MXFP4 · 11.81 GB · about 45.1 tokens/s

Best for code

Qwen2.5 Coder 14B Instruct

Q8_0 · 16.53 GB · about 12.1 tokens/s

Largest that still runs well

GPT OSS 20B

20.91B parameters · MXFP4 · about 45.1 tokens/s

Fastest useful answer

LFM2.5 8B A1B

about 65.3 tokens/s · 8.78 GB

See how fast it feels

Apple M4 Pro 24GB running GPT OSS 20B at MXFP4

Wait for the first word1.6 s
Then writes at45.1 tok/s
Whole answer5.4 s

YouWhy does my model use more memory when the conversation gets longer?

Model

Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.

Where it sits

The GPU can use about 16.8 GB of this machine's 24 GB by default, so for planning it behaves like a 16 GB card with different bandwidth. The 16 GB guide explains what that tier runs and where it runs out, and the bandwidth guide explains why 273 GB/s decides how fast it answers.

Our indicative price for this machine is $1,399 new, reviewed 2026-09-01. That is 1.7 GB of memory and 20 GB/s of bandwidth for every $100. Prices move, so treat these as a place to start a search and not a quote.

Devices in a similar price band, fastest memory first:

DevicePriceMemoryBandwidthBest all-round pick
Apple M1 Max 32GB $1,400 32 GB 400 GB/s Gemma 4 12B Instruct, 20.7 t/s
Ryzen AI Max+ 395 32GB $1,299 32 GB 256 GB/s GPT OSS 20B, 35.3 t/s
Ryzen AI Max+ 395 64GB $1,699 64 GB 256 GB/s Qwen3.5 35B A3B, 27.7 t/s
Apple M5 16GB $1,599 16 GB 153 GB/s Qwen3.5 9B, 11.8 t/s

Every model, scored on this device

Each row uses the highest-quality quantization that both fits and stays conversational. Speed is a single-stream estimate at 8k context.

Model Params Verdict Download Memory used Speed Max context
GPT OSS 20B
GPT-OSS 		20.91B (4.18B active) Runs great MXFP4 11.81 GB 45.1 t/s 128k
Qwen3.5 9B
Qwen 		9.65B Runs well Q8_0 10.22 GB 21 t/s 128k
Gemma 4 12B Instruct
Gemma 		11.96B Runs well Q8_0 13.52 GB 14.1 t/s 32k
Gemma 4 E4B Instruct
Gemma 		8B Runs well Q8_0 8.39 GB 23 t/s 128k
Qwen3 8B
Qwen 		8.19B Runs well Q8_0 9.61 GB 21.5 t/s 32k
Granite 4.1 8B
Granite 		8.79B Runs well Q8_0 10.32 GB 18.7 t/s 32k
Qwen2.5 VL 7B Instruct
Qwen 		8.29B Runs well Q8_0 9 GB 22.9 t/s 64k
Qwen3 14B
Qwen 		14.77B Just fits Q8_0 16.28 GB 12.3 t/s 8k
Qwen2.5 14B Instruct
Qwen 		14.77B Just fits Q8_0 16.53 GB 12.1 t/s 8k
Qwen2.5 7B Instruct
Qwen 		7.62B Runs well Q8_0 8.34 GB 25 t/s 32k
Gemma 3 12B Instruct
Gemma 		12.19B Runs well Q8_0 13.24 GB 14.4 t/s 64k
Llama 3.1 8B Instruct
Llama 		8.03B Runs well Q8_0 9.32 GB 22 t/s 64k
Qwen2.5 Coder 14B Instruct
Qwen 		14.77B Just fits Q8_0 16.53 GB 12.1 t/s 8k
DeepSeek Coder V2 Lite Instruct
DeepSeek 		15.71B (2.74B active) Just fits Q8_0 16.09 GB 37 t/s 16k
Phi 4
Phi 		14.66B Just fits Q8_0 16.48 GB 11.9 t/s 8k
Qwen2.5 Coder 7B Instruct
Qwen 		7.62B Runs well Q8_0 8.34 GB 25 t/s 32k
Mistral Nemo Instruct 2407
Mistral 		12.25B Runs well Q8_0 13.78 GB 14.6 t/s 16k
OLMo 3 7B Instruct
OLMo 		7.3B Runs well Q8_0 10.1 GB 19.9 t/s 32k
Internlm3 8B Instruct
InternLM 		8.8B Runs well Q8_0 9.46 GB 21.7 t/s 32k
Mistral 7B Instruct V0.3
Mistral 		7.25B Runs well Q8_0 8.55 GB 23.1 t/s 32k
Gemma 2 9B Instruct
Gemma 		9.24B Runs well Q8_0 11.47 GB 16.7 t/s 8k
Gemma 4 E2B Instruct
Gemma 		5.12B Runs great Q8_0 5.41 GB 36.3 t/s 128k
Qwen3.5 4B
Qwen 		4.66B Runs great Q8_0 5.24 GB 37.8 t/s 256k
Mistral 7B Instruct V0.2
Mistral 		7.24B Runs well Q8_0 8.54 GB 23.1 t/s 32k
DeepSeek Coder 7B Instruct V1.5
DeepSeek 		6.91B Runs well Q8_0 10.96 GB 18.2 t/s 4k
LFM2.5 8B A1B
Liquid 		8.47B (1.57B active) Runs great Q8_0 8.78 GB 65.3 t/s 64k
Qwen3 4B
Qwen 		4.02B Runs great Q8_0 5.43 GB 36.4 t/s 32k
DeepSeek Coder 6.7B Instruct
DeepSeek 		6.74B Runs well Q8_0 11.04 GB 17.6 t/s 16k
Gemma 3 4B Instruct
Gemma 		4.3B Runs great Q8_0 4.88 GB 40.8 t/s 128k
Phi 4 Mini Instruct
Phi 		3.84B Runs great Q8_0 5.14 GB 38.7 t/s 64k
Show the other 56 models
Granite 4.1 3B 3.4B Runs great Q8_0 4.31 GB 46.5 t/s 128k
Qwen2.5 3B Instruct 3.09B Runs great Q8_0 3.65 GB 55.6 t/s 32k
Phi 3 Mini 4k Instruct 3.82B Runs great Q8_0 4.87 GB 41.9 t/s 4k
Llama 3.2 3B Instruct 3.21B Runs great Q8_0 4.39 GB 45.8 t/s 64k
LFM2.5 2.6B 2.7B Runs great Q8_0 3.11 GB 66.4 t/s 128k
Qwen3.5 2B 2.27B Runs great Q8_0 2.67 GB 78.7 t/s 256k
Gemma 2 2B Instruct 2.61B Runs great Q8_0 3.51 GB 58.1 t/s 8k
Starcoder2 3B 3.03B Runs great Q8_0 3.46 GB 62.6 t/s 16k
Qwen3 1.7B 2.03B Runs great Q8_0 3.19 GB 64.4 t/s 32k
DeepSeek R1 Distill Qwen 1.5B 1.78B Runs great Q8_0 2.27 GB 106.1 t/s 128k
Qwen2.5 1.5B Instruct 1.54B Runs great Q8_0 2.03 GB 106.5 t/s 32k
SmolLM2 1.7B 1.71B Runs great Q8_0 3.5 GB 58.1 t/s 8k
OLMo 2 0425 1B 1.48B Runs great Q8_0 2.77 GB 82.1 t/s 4k
Llama 3.2 1B Instruct 1.24B Runs great Q8_0 1.79 GB 125.7 t/s 128k
LFM2.5 1.2B Instruct 1.17B Runs great Q8_0 1.56 GB 165.9 t/s 64k
Gemma 3 1B Instruct 1B Runs great Q8_0 1.32 GB 178.7 t/s 32k
Qwen3.5 0.8B 0.87B Runs great Q8_0 1.25 GB 190.8 t/s 256k
Qwen3 0.6B 0.75B Runs great Q8_0 1.89 GB 114.8 t/s 32k
Qwen2.5 0.5B Instruct 0.49B Runs great Q8_0 0.85 GB 320.8 t/s 32k
LFM2.5 350M 0.35B Runs great Q8_0 0.71 GB 496.7 t/s 64k
Qwen3 Next 80B A3B Instruct 81.32B Fits only at 2 to 3 bits n/a 20.07 GB 105.4 t/s n/a
Qwen3 Coder Next 79.67B Fits only at 2 to 3 bits n/a 19.67 GB 105.4 t/s n/a
Qwen3.5 35B A3B 35.95B Just fits n/a 20.74 GB 59 t/s n/a
GLM 4.7 Flash 31.22B Just fits n/a 21.4 GB 38.3 t/s n/a
Qwen3 30B A3B 30.53B Just fits n/a 21.28 GB 36.6 t/s n/a
Qwen3 Coder 30B A3B Instruct 30.53B Just fits n/a 21.28 GB 36.6 t/s n/a
Gemma 4 26B A4B Instruct 25.81B Just fits n/a 20.87 GB 29.3 t/s n/a
Phi 3.5 MoE Instruct 41.87B Just fits n/a 20.43 GB 26.8 t/s n/a
Mixtral 8x7B Instruct V0.1 46.7B Fits only at 2 to 3 bits n/a 18.01 GB 19.3 t/s n/a
Mistral Small 24B Instruct 2501 23.57B Just fits n/a 17.27 GB 11.3 t/s n/a
Gemma 3 27B Instruct 27.43B Just fits n/a 16.93 GB 11.2 t/s n/a
Gemma 4 31B Instruct 31.27B Just fits n/a 17.93 GB 10.6 t/s n/a
Qwen3 32B 32.76B Just fits n/a 18.62 GB 10.4 t/s n/a
Qwen2.5 32B Instruct 32.76B Just fits n/a 18.62 GB 10.4 t/s n/a
Qwen2.5 Coder 32B Instruct 32.76B Just fits n/a 18.62 GB 10.4 t/s n/a
Granite 4.1 30B 28.87B Just fits n/a 18.61 GB 10.2 t/s n/a
Qwen3.5 27B 27.78B Just fits n/a 19.46 GB 10.2 t/s n/a
Codestral 22B V0.1 22.25B Just fits n/a 19.18 GB 10 t/s n/a
Llama 3.3 70B Instruct 70.55B Fits only at 2 to 3 bits n/a 19.93 GB 9.7 t/s n/a
Qwen2.5 72B Instruct 72.71B Fits only at 2 to 3 bits n/a 20.45 GB 9.5 t/s n/a
Qwen2.5 VL 72B Instruct 73.41B Fits only at 2 to 3 bits n/a 20.62 GB 9.4 t/s n/a
Kimi K2.6 1026.88B Will not run n/a 578.41 GB n/a n/a
Kimi K2 Instruct 1026.41B Will not run n/a 578.15 GB n/a n/a
GLM 5.1 753.86B Will not run n/a 425.01 GB n/a n/a
GLM 5.2 753.33B Will not run n/a 424.72 GB n/a n/a
DeepSeek V3.2 685.4B Will not run n/a 386.4 GB n/a n/a
DeepSeek R1 684.53B Will not run n/a 385.92 GB n/a n/a
MiniMax M3 427.04B Will not run n/a 241.5 GB n/a n/a
GLM 4.5 358.34B Will not run n/a 204.77 GB n/a n/a
Qwen3 235B A22B 235.09B Will not run n/a 134.03 GB n/a n/a
MiniMax M2.5 228.7B Will not run n/a 130.88 GB n/a n/a
MiniMax M2.7 228.69B Will not run n/a 130.87 GB n/a n/a
Qwen3.5 122B A10B 125.09B Will not run n/a 71.01 GB n/a n/a
GPT OSS 120B 116.83B Will not run n/a 59.65 GB n/a n/a
GLM 4.5 Air 110.47B Will not run n/a 63.93 GB n/a n/a
Llama 3 3 Nemotron Super 49B V1 49.87B Will not run n/a 48.55 GB n/a n/a

What it will not run

36 of the 86 architectures we track are out of reach here, even at 2-bit precision.

If you want more headroom

The next steps up in memory, in order. More memory changes which models load at all; more bandwidth changes how fast they answer, so the two are worth weighing separately.

DeviceMemoryBandwidthModels that fit
GeForce RTX 5090 32 GB 1792 GB/s 69 Check price
Ryzen AI Max+ 395 32GB 32 GB 256 GB/s 55 Check price
Jetson AGX Orin 32GB 32 GB 204.8 GB/s 56 Check price
Intel Core Ultra 9 288V 32GB 32 GB 136 GB/s 53 Check price

Price links go to an Amazon search for the model name and are affiliate links: if you buy through one, we earn a commission at no cost to you. We do not take payment for placement, and the ordering above is by memory capacity alone.

Other devices with 24 GB

Same capacity, different speed. Once a model fits, bandwidth is what separates these.

GeForce RTX 4090

1008 GB/s · NVIDIA

GeForce RTX 3090 Ti

1008 GB/s · NVIDIA

Radeon RX 7900 XTX

960 GB/s · AMD

GeForce RTX 3090

936 GB/s · NVIDIA

GeForce RTX 5090 Laptop

896 GB/s · NVIDIA

RTX A5000

768 GB/s · NVIDIA

Questions people ask about an Apple M4 Pro 24GB

How much VRAM does an Apple M4 Pro 24GB have?

The Apple M4 Pro 24GB has 24 GB of unified memory shared with the CPU, of which roughly 16.8 GB is available to a model once the operating system takes its share.

Can an Apple M4 Pro 24GB run a 70B model?

Yes, once the macOS GPU memory limit is raised. At IQ2_XXS it needs 19.67 GB and runs at around 105.4 tokens per second.

What is the best local model for an Apple M4 Pro 24GB?

GPT OSS 20B at MXFP4 is the strongest all-round fit: 11.81 GB used, around 45.1 tokens per second at 8k context.