Hardware

Apple M4 Max (40-core GPU)

Apple M4 Max (40-core GPU) has 48 / 64 / 128 GB Unified and 546 GB/s. See which open models fit, at which quant, and roughly how fast. Figures below use the default memory (48 GB). Updated 2026-10-17.

Memory48 / 64 / 128 GB Unified

Bandwidth546 GB/s

Kindmac

Year2024

Apple: up to 546 GB/s and 128GB for the 40-core GPU M4 Max. Source.

RSS for new models on this device

Models that fit

Model	Verdict	Quant	Context	tok/s
Qwen3 Coder Next 79.7B	Tight fit	UD-Q3_K_XL	8K	43–59
Qwen3 Next 80B A3B Instruct 79.7B	Tight fit	UD-Q3_K_XL	8K	140–191
Qwen3 Next 80B A3B Thinking 79.7B	Tight fit	UD-Q3_K_XL	8K	140–191
Mistral Small 4 119B 2603 119B	Tight fit	UD-IQ2_XXS	8K	46–63
Qwen3.8 27B 27.3B	Runs	UD-Q8_K_L	8K	11–14
Kwaipilot KAT Coder V2.5 Dev 34.7B	Tight fit	Q8_0	8K	44–60
Qwen3.8 35B A3B Distill 35.5B	Runs	Q6_K	8K	114–155
Qwen3.6 35B A3B 34.7B	Tight fit	Q8_0	8K	89–122
Qwen3.5 35B A3B 34.7B	Tight fit	Q8_0	8K	89–122
Llama 4 Scout 17B 16E Instruct 108B	Tight fit	UD-IQ1_M	8K	33–45
NVIDIA Nemotron 3.5 Lightning 30B A3B 32.9B	Tight fit	Q8_0	8K	83–113
Devstral Small 2 24B Instruct 2512 23.6B	Runs	UD-Q8_K_XL	8K	9.9–13
Mistral Small 3.2 24B Instruct 2506 23.6B	Runs	UD-Q8_K_XL	8K	9.9–13
Magistral Small 2509 23.6B	Runs	UD-Q8_K_XL	8K	9.9–13
Nemotron 3 Nano 30B A3B 31.6B	Tight fit	Q8_0	8K	83–113
NVIDIA Nemotron 3 Nano Omni 30B A3B Reasoning 31.6B	Tight fit	UD-Q8_K_XL	8K	79–108
GLM 4.7 Flash 29.9B	Tight fit	UD-Q8_K_XL	8K	37–51
Qwen3 Coder 30B A3B Instruct 30.5B	Tight fit	UD-Q8_K_XL	8K	69–94
Qwen3 30B A3B Instruct 2507 30.5B	Tight fit	UD-Q8_K_XL	8K	69–94
Qwen3 VL 30B A3B Instruct 30.5B	Tight fit	UD-Q8_K_XL	8K	69–94
Qwen3 30B A3B Thinking 2507 30.5B	Tight fit	UD-Q8_K_XL	8K	69–94
gemma 4 26B A4B it 25.2B	Runs	UD-Q8_K_XL	8K	47–64
LFM2.5 24B A2B 24.0B	Runs	Q8_0	8K	107–146
GLM 4.7 Flash REAP 23B A3B 23.0B	Runs	UD-Q8_K_XL	8K	74–101
gpt oss 20b 20.9B	Runs great	UD-Q8_K_XL	8K	74–102
Qwen2.5 14B Instruct 14.8B	Runs	F16	8K	9.6–13
Qwen3 14B 14.8B	Runs	BF16	8K	9.7–13
NousResearch Hermes 4 14B 14.8B	Runs	BF16	8K	9.7–13
DeepSeek R1 Distill Qwen 14B 14.8B	Runs	F16	8K	9.6–13
Qwen2.5 Coder 14B Instruct 128K 14.8B	Runs	F16	8K	9.6–13
phi 4 14.7B	Runs	F16	8K	9.7–13
Ministral 3 14B Instruct 2512 13.5B	Runs	BF16	8K	11–14
Ministral 3 14B Reasoning 2512 13.5B	Runs	BF16	8K	11–14
gemma 4 12b it 11.9B	Runs	BF16	8K	11–15
GLM 4.6V Flash 9.4B	Runs great	BF16	8K	16–21
GLM Z1 9B 0414 9.4B	Runs great	BF16	8K	16–21
GLM 4 9B 0414 9.4B	Runs great	BF16	8K	16–21
GLM 4.1V 9B Thinking 9.4B	Runs great	BF16	8K	16–21
Qwen3.5 9B 9.0B	Runs great	BF16	8K	17–23
MiMo V2.6 Distill Qwen 9B 9.0B	Runs great	BF16	8K	17–23

90 more models in the catalog do not fit this default profile, or ranked below the first 40 that do.

Example: Tight fit fully in unified memory on your Apple M4 Max (40-core GPU) at UD-Q3_K_XL, 8K context, about 43–59 tok/s. Weights 33.8 GB plus 0.75 GB of KV cache — 35.2 GB of 36.0 GB usable. Weights shown in the verdict use GiB (the same 1024³ counting as GPU memory).