Which local LLMs can the Apple M4 (Mac mini, MacBook Air, MacBook Pro, iMac) run?

Unified memory — by default the GPU can use about 70% of it

Assumes 8k context and an f16 KV cache. Verdicts are shown for each memory size.

What to expect from the Apple M4

The base M4 shares its memory between CPU and GPU, and CanRun lets the GPU use only part of it: on the 32GB model that is 22.4 GB. The bigger limit is speed. At 120 GB/s of memory bandwidth, against 273 GB/s on the Apple M4 Pro, the M4 reads model weights slowly, so the memory size you choose decides what loads, not how fast it runs.

With 16GB, Llama 3.1 8B fits and runs well. Qwen3 8B runs well too, even though it has a thinking mode and CanRun holds such models to a 1.5× stricter speed bar. Llama 3.2 3B runs great. gpt-oss-20b needs the 24GB model and then runs great.

With 32GB, mixture-of-experts models do best: Qwen3 30B-A3B runs great and Gemma 4 26B-A4B runs well, while dense models such as Gemma 4 31B fit but run slowly. Qwen3 32B is too big for the GPU's share at Q4_K_M and would run on the CPU alone; its IQ4_XS file fits on the GPU and runs slowly. Our Mac guide compares the M4 and M5 chips side by side.

Specs

Memory options: 16 GB · 24 GB · 32 GB
Memory bandwidth: 120.0 GB/s
FP16 compute: 3.8 TFLOPS
Launch year: 2024
Macs with this chip: Mac mini: 16 GB · 24 GB · 32 GB
MacBook Air: 16 GB · 24 GB · 32 GB
MacBook Pro: 16 GB · 24 GB · 32 GB
iMac: 16 GB · 24 GB · 32 GB
Also written as: M4

Verdicts at a glance

At Q4_K_M (or the closest available quant) with 8k context.

16 GBRuns great11 modelsRuns well12 modelsRuns slowly5 modelsWon't run31 models5 more models run at a lower quant.
24 GBRuns great13 modelsRuns well13 modelsRuns slowly14 modelsWon't run19 models8 more models run at a lower quant.
32 GBRuns great15 modelsRuns well14 modelsRuns slowly17 modelsWon't run13 models4 more models run at a lower quant.

How to read the verdicts

Runs great: Fully on the GPU at 20 tok/s or more
Runs well: Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
Runs slowly: 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
Won't run: Does not fit, or under 2 tok/s

Every model on the Apple M4

Scroll sideways to see every column.

59 models sorted by verdict and speed, with 32 GB of memory
Model	Verdict	Speed	Quant	Memory	Runs as	Context	Notes
MiniCPM5 1B	Runs great	est. 102.3 tok/s81.9–122.8calibrated estimate ±20%	Q4_K_M	1.5 / 22.4 GB	Unified memory	up to 64k	thinking mode
Ling-3.0 Tiny	Runs great	est. 93.7 tok/s65.6–121.9theoretical estimate ±30%	Q4_K_M	5.5 / 22.4 GB	Unified memory	up to 128k	thinking mode
Llama 3.2 1B	Runs great	est. 84.6 tok/s67.7–101.5calibrated estimate ±20%	Q4_K_M	1.7 / 22.4 GB	Unified memory	up to 64k	—
EXAONE 4.0 1.2B	Runs great	est. 69.4 tok/s55.6–83.3calibrated estimate ±20%	Q4_K_M	1.9 / 22.4 GB	Unified memory	up to 32k	thinking mode
Gemma 4 E2B	Runs great	est. 53.3 tok/s37.3–69.3theoretical estimate ±30%	Q4_K_M	3.7 / 22.4 GB	Unified memory	up to 128k	thinking mode
Mellum2.1 12B-A2.5B Thinking	Runs great	est. 51.2 tok/s40.9–61.4calibrated estimate ±20%	Q4_K_M	8.8 / 22.4 GB	Unified memory	up to 64k	thinking mode
LFM2.5 2.6B	Runs great	est. 50.5 tok/s35.4–65.7theoretical estimate ±30%	Q4_K_M	2.4 / 22.4 GB	Unified memory	up to 64k	thinking mode
HyperCLOVA X SEED 1.5B	Runs great	est. 50.2 tok/s40.2–60.3calibrated estimate ±20%	Q4_K_M	2.4 / 22.4 GB	Unified memory	up to 16k	—
MiniCPM5 2B	Runs great	est. 47.7 tok/s38.2–57.2calibrated estimate ±20%	Q4_K_M	2.5 / 22.4 GB	Unified memory	up to 32k	thinking mode
MiniCPM-V 4.7 35B-A3B	Runs great	est. 37.5 tok/s30.0–45.0calibrated estimate ±20%	Q4_K_M	21.9 / 22.4 GB	Unified memory	up to 16k	thinking mode
gpt-oss-20b	Runs greatgpt-oss-20b verdict on the Apple M4	est. 31.9 tok/s25.5–38.3calibrated estimate ±20%	MXFP4	12.9 / 22.4 GB	Unified memory	up to 8k	thinking mode
Spark-X2.5 4B	Runs great	est. 31.4 tok/s25.1–37.7calibrated estimate ±20%	Q4_K_M	3.5 / 22.4 GB	Unified memory	up to 8k	thinking mode
Llama 3.2 3B	Runs great	est. 30.8 tok/s24.7–37.0calibrated estimate ±20%	Q4_K_M	3.5 / 22.4 GB	Unified memory	up to 16k	—
Qwen3 30B-A3B (2507)	Runs greatQwen3 30B-A3B (2507) verdict on the Apple M4	est. 29.8 tok/s23.8–35.7calibrated estimate ±20%	Q4_K_M	19.9 / 22.4 GB	Unified memory	up to 16k	—
Kanana 1.5 15.7B-A3B	Runs great	est. 29.1 tok/s23.3–35.0calibrated estimate ±20%	Q4_K_M	12.1 / 22.4 GB	Unified memory	up to 16k	—
Qwen3.5 4B	Runs well	est. 29.4 tok/s23.5–35.3calibrated estimate ±20%	Q4_K_M	3.7 / 22.4 GB	Unified memory	up to 128k	Try IQ4_XS: Runs great thinking mode
Gemma 4 E4B	Runs well	est. 27.4 tok/s19.1–35.6theoretical estimate ±30%	Q4_K_M	5.7 / 22.4 GB	Unified memory	up to 128k	UD-Q2_K_XL (heavy quality loss): Runs great thinking mode
Gemma 4 26B-A4B	Runs wellGemma 4 26B-A4B verdict on the Apple M4	est. 23.6 tok/s18.9–28.4calibrated estimate ±20%	Q4_K_M	17.7 / 22.4 GB	Unified memory	up to 128k	UD-Q2_K_XL (heavy quality loss): Runs great thinking mode
DeepSeek R1 Distill Llama 8B	Runs well	est. 15.2 tok/s12.2–18.3calibrated estimate ±20%	Q4_K_M	6.6 / 22.4 GB	Unified memory	up to 16k	thinking mode
Kanana 1.5 8B	Runs well	est. 15.2 tok/s12.2–18.3calibrated estimate ±20%	Q4_K_M	6.6 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs great
Llama 3.1 8B	Runs wellLlama 3.1 8B verdict on the Apple M4	est. 15.2 tok/s12.2–18.3calibrated estimate ±20%	Q4_K_M	6.6 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs great
Nanbeige4.2 3B	Runs well	est. 14.9 tok/s10.4–19.4theoretical estimate ±30%	Q4_K_M	4.7 / 22.4 GB	Unified memory	up to 8k	thinking mode
Qwen3.5 9B	Runs wellQwen3.5 9B verdict on the Apple M4	est. 14.9 tok/s11.9–17.8calibrated estimate ±20%	Q4_K_M	6.7 / 22.4 GB	Unified memory	up to 32k	thinking mode
DeepSeek R1 0528 Qwen3 8B	Runs well	est. 14.6 tok/s11.7–17.5calibrated estimate ±20%	Q4_K_M	6.8 / 22.4 GB	Unified memory	up to 16k	thinking mode
Qwen3 8B	Runs wellQwen3 8B verdict on the Apple M4	est. 14.6 tok/s11.7–17.5calibrated estimate ±20%	Q4_K_M	6.8 / 22.4 GB	Unified memory	up to 16k	thinking mode
K2-Horizon 7B	Runs well	est. 13.4 tok/s10.7–16.1calibrated estimate ±20%	Q4_K_M	7.4 / 22.4 GB	Unified memory	up to 8k	thinking mode
Gemma 4 12B	Runs wellGemma 4 12B verdict on the Apple M4	est. 12.6 tok/s10.1–15.1calibrated estimate ±20%	Q4_K_M	7.8 / 22.4 GB	Unified memory	up to 16k	thinking mode
Gemma 3 12B	Runs wellGemma 3 12B verdict on the Apple M4	est. 11.6 tok/s9.3–14.0calibrated estimate ±20%	Q4_K_M	8.4 / 22.4 GB	Unified memory	up to 32k	—
Phi-4	Runs well	est. 8.5 tok/s6.8–10.2calibrated estimate ±20%	Q4_K_M	11.3 / 22.4 GB	Unified memory	up to 8k	—
Qwen3.6 35B-A3B	Runs slowlyQwen3.6 35B-A3B verdict on the Apple M4	est. 25.4 tok/s17.8–33.1theoretical estimate ±30%	Q4_K_M	22.5 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly Try IQ4_XS: Runs great thinking mode
Qwen3.5 35B-A3B	Runs slowlyQwen3.5 35B-A3B verdict on the Apple M4	est. 25.1 tok/s17.6–32.6theoretical estimate ±30%	Q4_K_M	22.8 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly Try IQ4_XS: Runs great thinking mode
Nemotron 3.5 Lightning 30B-A3B	Runs slowly	est. 20.9 tok/s14.6–27.1theoretical estimate ±30%	Q4_K_M	25.5 GB RAM	CPU only	up to 256k	CPU inference — capped at Runs slowly Try IQ4_XS: Runs great thinking mode
K2-Horizon MoVA 36B-A4B	Runs slowly	est. 12.9 tok/s9.0–16.8theoretical estimate ±30%	Q4_K_M	24.0 GB RAM	CPU only	up to 16k	CPU inference — capped at Runs slowly thinking mode
HyperCLOVA X SEED Think 14B	Runs slowly	est. 8.9 tok/s6.3–11.6theoretical estimate ±30%	Q4_K_M	10.8 / 22.4 GB	Unified memory	up to 64k	thinking mode
Qwen3 14B	Runs slowlyQwen3 14B verdict on the Apple M4	est. 8.8 tok/s7.1–10.6calibrated estimate ±20%	Q4_K_M	10.9 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well thinking mode
DeepSeek R1 Distill Qwen 14B	Runs slowly	est. 8.6 tok/s6.9–10.3calibrated estimate ±20%	Q4_K_M	11.2 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well thinking mode
Mistral Small 3.2 24B	Runs slowly	est. 5.8 tok/s4.7–7.0calibrated estimate ±20%	Q4_K_M	16.2 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well
Gemma 3 27B	Runs slowly	est. 5.3 tok/s4.2–6.4calibrated estimate ±20%	Q4_K_M	17.8 / 22.4 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well
Qwen3.6 27B	Runs slowly	est. 5.3 tok/s4.2–6.3calibrated estimate ±20%	Q4_K_M	17.9 / 22.4 GB	Unified memory	up to 64k	thinking mode
Muse Glimmer 30B	Runs slowly	est. 5.2 tok/s4.2–6.3calibrated estimate ±20%	Q4_K_M	18.0 / 22.4 GB	Unified memory	up to 128k	thinking mode
Qwen3.5 27B	Runs slowlyQwen3.5 27B verdict on the Apple M4	est. 5.2 tok/s4.1–6.2calibrated estimate ±20%	Q4_K_M	18.2 / 22.4 GB	Unified memory	up to 32k	thinking mode
Qwen3.8 27B	Runs slowlyQwen3.8 27B verdict on the Apple M4	est. 5.1 tok/s4.1–6.1calibrated estimate ±20%	Q4_K_M	18.6 / 22.4 GB	Unified memory	up to 32k	thinking mode
Gemma 4 31B	Runs slowlyGemma 4 31B verdict on the Apple M4	est. 4.8 tok/s3.8–5.8calibrated estimate ±20%	Q4_K_M	19.6 / 22.4 GB	Unified memory	up to 32k	thinking mode
Granite 4.2 30B	Runs slowly	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	Q4_K_M	20.4 / 22.4 GB	Unified memory	up to 8k	thinking mode
EXAONE 4.0 32B	Runs slowly	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	Q4_K_M	20.5 / 22.4 GB	Unified memory	up to 32k	thinking mode
EXAONE 4.5 33B	Runs slowly	est. 4.4 tok/s3.5–5.3calibrated estimate ±20%	Q4_K_M	21.2 / 22.4 GB	Unified memory	up to 16k	thinking mode
DeepSeek R1 Distill Qwen 32B	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 6.3 tok/s5.0–7.6calibrated estimate ±20%	Q2_K	15.0 / 22.4 GB	Unified memory	up to 32k	thinking mode
Qwen3 32B	Won't runTry IQ4_XS: Runs slowlyQwen3 32B verdict on the Apple M4	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	IQ4_XS	20.4 / 22.4 GB	Unified memory	up to 8k	thinking mode
Nanbeige4.3 33B	Won't runTry IQ4_XS: Runs slowly	est. 4.5 tok/s3.6–5.4calibrated estimate ±20%	IQ4_XS	21.0 / 22.4 GB	Unified memory	up to 8k	thinking mode
Llama 3.3 70B	Won't runTry IQ2_M (heavy quality loss): Runs slowlyLlama 3.3 70B verdict on the Apple M4	est. 2.2 tok/s1.6–2.9theoretical estimate ±30%	IQ2_M	26.8 GB RAM	CPU only	up to 8k	—
DeepSeek V4 Flash 0731	Won't run	—	MXFP4	needs 155.6 GB	—	—	Needs about 155.6 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
GLM-5.3 Flash	Won't run	—	Q4_K_M	needs 182.1 GB	—	—	Needs about 182.1 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
gpt-oss-120b	Won't rungpt-oss-120b verdict on the Apple M4	—	MXFP4	needs 64.3 GB	—	—	Needs about 64.3 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Kolibri 1	Won't run	—	Q4_K_M	needs 50.5 GB	—	—	Needs about 50.5 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Laguna S 2.1	Won't run	—	Q4_K_M	needs 68.6 GB	—	—	Needs about 68.6 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Ling-3.0 Flash	Won't run	—	Q4_K_M	needs 77.7 GB	—	—	Needs about 77.7 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Qwen3.5 122B-A10B	Won't runQwen3.5 122B-A10B verdict on the Apple M4	—	Q4_K_M	needs 79.0 GB	—	—	Needs about 79.0 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Solar Open 100B	Won't run	—	Q4_K_M	needs 64.4 GB	—	—	Needs about 64.4 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode
Solar Open 2 250B	Won't run	—	IQ4_XS	needs 137.2 GB	—	—	Needs about 137.2 GB; 28.0 GB of memory is free (22.4 GB usable by the GPU) thinking mode

Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.

Verdict by memory size

Q4_K_M verdict for each memory configuration — open the model page for speeds
Model	16 GB	24 GB	32 GB
MiniCPM5 1B	Runs great	Runs great	Runs great
Ling-3.0 Tiny	Runs great	Runs great	Runs great
Llama 3.2 1B	Runs great	Runs great	Runs great
EXAONE 4.0 1.2B	Runs great	Runs great	Runs great
Gemma 4 E2B	Runs great	Runs great	Runs great
Mellum2.1 12B-A2.5B Thinking	Runs great	Runs great	Runs great
LFM2.5 2.6B	Runs great	Runs great	Runs great
HyperCLOVA X SEED 1.5B	Runs great	Runs great	Runs great
MiniCPM5 2B	Runs great	Runs great	Runs great
MiniCPM-V 4.7 35B-A3B	Won't run	Won't run	Runs great
gpt-oss-20b	Won't run	Runs great	Runs great
Spark-X2.5 4B	Runs great	Runs great	Runs great
Llama 3.2 3B	Runs great	Runs great	Runs great
Qwen3 30B-A3B (2507)	Won't run	Runs slowly	Runs great
Kanana 1.5 15.7B-A3B	Runs slowly	Runs great	Runs great
Qwen3.5 4B	Runs well	Runs well	Runs well
Gemma 4 E4B	Runs well	Runs well	Runs well
Gemma 4 26B-A4B	Won't run	Runs slowly	Runs well
DeepSeek R1 Distill Llama 8B	Runs well	Runs well	Runs well
Kanana 1.5 8B	Runs well	Runs well	Runs well
Llama 3.1 8B	Runs well	Runs well	Runs well
Nanbeige4.2 3B	Runs well	Runs well	Runs well
Qwen3.5 9B	Runs well	Runs well	Runs well
DeepSeek R1 0528 Qwen3 8B	Runs well	Runs well	Runs well
Qwen3 8B	Runs well	Runs well	Runs well
K2-Horizon 7B	Runs well	Runs well	Runs well
Gemma 4 12B	Runs well	Runs well	Runs well
Gemma 3 12B	Runs well	Runs well	Runs well
Phi-4	Runs slowly	Runs well	Runs well
Qwen3.6 35B-A3B	Won't run	Won't run	Runs slowly
Qwen3.5 35B-A3B	Won't run	Won't run	Runs slowly
Nemotron 3.5 Lightning 30B-A3B	Won't run	Won't run	Runs slowly
K2-Horizon MoVA 36B-A4B	Won't run	Won't run	Runs slowly
HyperCLOVA X SEED Think 14B	Runs slowly	Runs slowly	Runs slowly
Qwen3 14B	Runs slowly	Runs slowly	Runs slowly
DeepSeek R1 Distill Qwen 14B	Runs slowly	Runs slowly	Runs slowly
Mistral Small 3.2 24B	Won't run	Runs slowly	Runs slowly
Gemma 3 27B	Won't run	Runs slowly	Runs slowly
Qwen3.6 27B	Won't run	Runs slowly	Runs slowly
Muse Glimmer 30B	Won't run	Runs slowly	Runs slowly
Qwen3.5 27B	Won't run	Runs slowly	Runs slowly
Qwen3.8 27B	Won't run	Runs slowly	Runs slowly
Gemma 4 31B	Won't run	Runs slowly	Runs slowly
Granite 4.2 30B	Won't run	Runs slowly	Runs slowly
EXAONE 4.0 32B	Won't run	Runs slowly	Runs slowly
EXAONE 4.5 33B	Won't run	Won't run	Runs slowly
DeepSeek R1 Distill Qwen 32B	Won't run	Won't run	Won't run
Qwen3 32B	Won't run	Won't run	Won't run
Nanbeige4.3 33B	Won't run	Won't run	Won't run
Llama 3.3 70B	Won't run	Won't run	Won't run
DeepSeek V4 Flash 0731	Won't run	Won't run	Won't run
GLM-5.3 Flash	Won't run	Won't run	Won't run
gpt-oss-120b	Won't run	Won't run	Won't run
Kolibri 1	Won't run	Won't run	Won't run
Laguna S 2.1	Won't run	Won't run	Won't run
Ling-3.0 Flash	Won't run	Won't run	Won't run
Qwen3.5 122B-A10B	Won't run	Won't run	Won't run
Solar Open 100B	Won't run	Won't run	Won't run
Solar Open 2 250B	Won't run	Won't run	Won't run

Measured results on the Apple M4

Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.

Measured results on the Apple M4
Model	Quant	Backend	Context	Prompt (tok/s)	Generation (tok/s)	Flags	Source	Measured
llama-2-7b	Q4_0	llama.cpp	512	221.0	24.1	Metal, b1550 (2023)	github.com	2024-11-10

Frequently asked questions

What is the largest model that runs entirely on the Apple M4?

Qwen3.5 35B-A3B at IQ4_XS (a 18.17 GB file) fits entirely in 32 GB of unified memory with 8k context, at est. 31.0 tok/s (24.8–37.2, calibrated estimate ±20%).

How many local LLMs run well on the Apple M4?

At Q4_K_M with 8k context, with 32 GB of memory, out of 59 tracked models: 15 run great, 14 run well, 17 run slowly and 13 won't run.

Can the Apple M4 run a 70B model like Llama 3.3 70B?

Runs slowly — CPU only, IQ2_M (heavy quality loss): est. 2.2 tok/s (1.6–2.9, theoretical estimate ±30%).

Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.