What are the best models to run on Apple M5 Max (48GB)?

Based on model rankings and 36 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~30 tok/s. Top Coding: Qwen3.5-27B (INT4) ~30 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~40 tok/s. Maximum Model Size: DeepSeek-R1 70B (INT4) ~12 tok/s.

How are VRAM, throughput, and latency calculated?

Calculations are determined using apxml.com modeling, combining LLM architecture analysis with GPU hardware physics. VRAM and Context Scaling: Memory requirements are determined from model architecture, including parameter count, layer configuration, and attention structure. The KV cache footprint scales deterministically based on context length, batch size, and cache precision. Throughput (TPS): Token generation is calculated in the memory-bandwidth-bound decode phase, measuring the time required to stream model weights and KV cache relative to hardware memory bandwidth. Latency (TTFT): Time to first token is calculated from the compute operations required during prompt prefill relative to hardware tensor compute capacity.

How accurate are these estimates?

Calculations are calibrated against empirical ground-truth benchmark runs across diverse architectures and hardware. Because the underlying mechanics are benchmark-calibrated, estimates remain accurate across different configurations without needing to run physical tests for every scenario. VRAM Accuracy: Memory estimates are deterministic and closely match actual allocation. Minor differences typically stay within 3% to 5% due to CUDA runtime overhead or memory allocator fragmentation. Speed and Latency Accuracy: Throughput and TTFT typically vary within 5% to 10% depending on the specific inference runtime, kernel implementations, and system power limits.

Best Models to Run on Apple M5 Max (48GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4 Q8	30 tok/s 17 tok/s	~16.3s ~17.7s
#48	Agnes 3.0 Flash	33B	INT4 Q8	26 tok/s 13 tok/s	~19.8s ~23.4s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4 Q8	110 tok/s 72 tok/s	~3.2s ~3.8s
#83	Qwen3.5-27B	27B	INT4 Q8	30 tok/s 17 tok/s	~16.3s ~17.7s
#92	Sarvam-30B	32B (2.4B active)	INT4 Q8	137 tok/s 102 tok/s	~2.2s ~2.4s
#94	Qwen3.6 35B A3B	35B (3B active)	INT4 Q8	125 tok/s 86 tok/s	~2.6s ~3.1s
#96	Muse Glimmer 30B	30B	INT4 Q8	28 tok/s 15 tok/s	~18.1s ~19.7s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8 FP16	40 tok/s 27 tok/s 15 tok/s	~8.6s ~8.6s ~9.4s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8 FP16	3 tok/s 2 tok/s 1 tok/s	~178.3s ~178.6s ~210.7s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4 Q8	117 tok/s 85 tok/s	~2.9s ~2.9s
#109	Gemma 4 12B	11.95B	INT4 Q8 FP16	45 tok/s 31 tok/s 18 tok/s	~7.3s ~7.4s ~7.4s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	75 tok/s 47 tok/s 26 tok/s	~5.6s ~5.6s ~5.6s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4 Q8	125 tok/s 86 tok/s	~2.6s ~3.1s
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	26 tok/s 16 tok/s 8 tok/s	~17.5s ~17.5s ~17.6s
#129	Gemma 4 31B	30.7B	INT4 Q8	27 tok/s 15 tok/s	~18.5s ~20.1s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4 Q8	55 tok/s 45 tok/s	~2.6s ~3.0s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	127 tok/s 88 tok/s 53 tok/s	~2.5s ~2.5s ~2.6s
#133	Qwen3-30B-A3B	30B (3B active)	INT4 Q8	127 tok/s 91 tok/s	~2.5s ~2.8s
#136	GPT-OSS 20B	21B (3.6B active)	INT4 Q8	61 tok/s 53 tok/s	~2.7s ~2.7s
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4 Q8	127 tok/s 91 tok/s	~2.5s ~2.8s
#153	Qwen3-32B	32B	INT4 Q8	26 tok/s 14 tok/s	~19.3s ~21.0s
#155	Mistral Large 3	41B (675B active)	INT4	1 tok/s	~423.1s
#156	GLM-4	32B	INT4	21 tok/s	~19.3s
#157	DeepSeek-R1 70B	70B	INT4	12 tok/s	~49.2s
#158	GLM-4V	9B	INT4 Q8 FP16	56 tok/s 39 tok/s 23 tok/s	~5.6s ~5.6s ~5.6s
#159	Gemma 3 27B	27B	INT4 Q8	30 tok/s 17 tok/s	~16.3s ~17.7s
#160	Phi-4	14B	INT4 Q8 FP16	53 tok/s 32 tok/s 16 tok/s	~8.6s ~8.6s ~9.4s
#161	Llama 3.3 70B	70B	INT4	11 tok/s	~49.2s
#162	DeepSeek-R1 32B	32B	INT4 Q8	27 tok/s 15 tok/s	~19.3s ~21.0s
#163	Hunyuan Standard	52B (389B active)	INT4	2 tok/s	~264.5s
#164	Qwen2.5-32B	32B	INT4 Q8	26 tok/s 14 tok/s	~19.3s ~21.0s
#168	Magistral Small	24B	INT4 Q8	26 tok/s 16 tok/s	~14.5s ~15.8s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.3s ~4.4s ~4.4s
#170	OLMo 3.1 32B Instruct	32B	INT4	21 tok/s	~19.3s
#171	Llama 3.1 70B	70B	INT4	11 tok/s	~49.2s
#172	DeepSeek-R1 14B	14B	INT4 Q8 FP16	55 tok/s 33 tok/s 17 tok/s	~8.6s ~8.6s ~9.4s
#173	Ministral 3 14B	14B	INT4 Q8 FP16	40 tok/s 27 tok/s 15 tok/s	~8.6s ~8.6s ~9.4s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	60 tok/s 37 tok/s 20 tok/s	~7.4s ~7.4s ~7.4s
#180	OLMo 3 32B Base	32B	INT4	21 tok/s	~19.3s
#181	Mistral-Small-2501	24B	INT4 Q8	34 tok/s 20 tok/s	~14.5s ~14.5s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	61 tok/s 43 tok/s 26 tok/s	~4.9s ~5.0s ~5.0s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	174 tok/s 134 tok/s 89 tok/s	~1.4s ~1.4s ~1.4s
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	174 tok/s 134 tok/s 89 tok/s	~1.4s ~1.4s ~1.4s
#185	Qwen2.5-14B	14B	INT4 Q8 FP16	53 tok/s 32 tok/s 16 tok/s	~8.6s ~8.6s ~9.4s
#186	Gemma 2 27B	27B	INT4 Q8	30 tok/s 17 tok/s	~16.3s ~17.7s
#187	OLMo 3.1 32B Think	32B	INT4	21 tok/s	~19.3s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	110 tok/s 74 tok/s 43 tok/s	~3.2s ~3.2s ~3.2s
#189	Qwen3-4B	4B	INT4 Q8 FP16	127 tok/s 88 tok/s 53 tok/s	~2.5s ~2.5s ~2.6s
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	84 tok/s 53 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#191	Llama 3 70B	70B	INT4	11 tok/s	~49.2s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	105 tok/s 84 tok/s 57 tok/s	~1.9s ~2.0s ~2.0s
#195	Qwen3-14B	14B	INT4 Q8 FP16	53 tok/s 32 tok/s 16 tok/s	~8.6s ~8.6s ~9.4s
#196	Kimi Linear 48B A3B Instruct	48B (3B active)	INT4	45 tok/s	~3.0s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	75 tok/s 47 tok/s 26 tok/s	~5.6s ~5.6s ~5.6s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
#199	Qwen3-8B	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	127 tok/s 88 tok/s 53 tok/s	~2.5s ~2.5s ~2.6s
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	92 tok/s 59 tok/s 33 tok/s	~4.3s ~4.3s ~4.4s
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	186 tok/s 146 tok/s 99 tok/s	~1.2s ~1.2s ~1.2s
#205	Command R	35B	INT4	19 tok/s	~21.0s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	259 tok/s 245 tok/s 218 tok/s	~184 ms ~184 ms ~184 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	249 tok/s 220 tok/s 176 tok/s	~408 ms ~408 ms ~409 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	130 tok/s 91 tok/s 55 tok/s	~2.4s ~2.4s ~2.4s
#211	Mixtral-8x7B-v0.1	46.7B (7B active)	INT4	81 tok/s	~5.0s
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	82 tok/s 52 tok/s 29 tok/s	~4.9s ~5.0s ~5.0s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	195 tok/s 156 tok/s 108 tok/s	~1.0s ~1.0s ~1.0s
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	147 tok/s 106 tok/s 66 tok/s	~1.9s ~1.9s ~2.0s
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	234 tok/s 201 tok/s 154 tok/s	~551 ms ~551 ms ~553 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	130 tok/s 91 tok/s 55 tok/s	~2.4s ~2.4s ~2.4s
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	105 tok/s 92 tok/s 68 tok/s	~1.4s ~1.4s ~1.5s
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	87 tok/s 56 tok/s 31 tok/s	~4.5s ~4.5s ~4.5s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	220 tok/s 185 tok/s 137 tok/s	~699 ms ~700 ms ~702 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	174 tok/s 134 tok/s 89 tok/s	~1.4s ~1.4s ~1.4s
#221	Phi-3-medium	14B	INT4 Q8 FP16	53 tok/s 32 tok/s 16 tok/s	~8.6s ~8.6s ~9.4s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.3s ~4.4s ~4.4s
#223	Yi-34B	34B	INT4	20 tok/s	~20.4s
#224	Qwen2-7B	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
#225	Phi-3-small	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	220 tok/s 185 tok/s 137 tok/s	~699 ms ~700 ms ~702 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	87 tok/s 56 tok/s 31 tok/s	~4.5s ~4.5s ~4.5s
#229	Gemma 1 7B	7B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.3s ~4.4s ~4.4s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	183 tok/s 139 tok/s 91 tok/s	~1.4s ~1.4s ~1.4s
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	73 tok/s 54 tok/s 33 tok/s	~3.7s ~3.8s ~3.8s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	73 tok/s 54 tok/s 33 tok/s	~3.7s ~3.8s ~3.8s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	73 tok/s 54 tok/s 33 tok/s	~3.7s ~3.8s ~3.8s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	257 tok/s 231 tok/s 189 tok/s	~338 ms ~339 ms ~340 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	257 tok/s 231 tok/s 189 tok/s	~338 ms ~339 ms ~340 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	254 tok/s 239 tok/s 210 tok/s	~204 ms ~204 ms ~205 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	254 tok/s 239 tok/s 210 tok/s	~204 ms ~204 ms ~205 ms
-	Falcon-1B	1B	INT4 Q8 FP16	228 tok/s 190 tok/s 139 tok/s	~699 ms ~700 ms ~702 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	220 tok/s 185 tok/s 137 tok/s	~699 ms ~700 ms ~702 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	195 tok/s 156 tok/s 108 tok/s	~1.0s ~1.0s ~1.0s
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	195 tok/s 156 tok/s 108 tok/s	~1.0s ~1.0s ~1.0s
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	186 tok/s 146 tok/s 99 tok/s	~1.2s ~1.2s ~1.2s
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	182 tok/s 142 tok/s 95 tok/s	~1.2s ~1.2s ~1.2s
-	Falcon-3B	3B	INT4 Q8 FP16	153 tok/s 109 tok/s 67 tok/s	~1.9s ~1.9s ~2.0s
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	151 tok/s 132 tok/s 99 tok/s	~893 ms ~894 ms ~896 ms
-	Phi-1	1.3B	INT4 Q8 FP16	151 tok/s 132 tok/s 99 tok/s	~893 ms ~894 ms ~896 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	151 tok/s 132 tok/s 99 tok/s	~893 ms ~894 ms ~896 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	148 tok/s 107 tok/s 66 tok/s	~1.9s ~1.9s ~2.0s
-	Falcon3-3B	3B	INT4 Q8 FP16	147 tok/s 106 tok/s 66 tok/s	~1.9s ~1.9s ~2.0s
-	Qwen2.5-3B	3B	INT4 Q8 FP16	147 tok/s 106 tok/s 66 tok/s	~1.9s ~1.9s ~2.0s
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	132 tok/s 99 tok/s	~2.4s ~2.4s
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	132 tok/s 99 tok/s	~2.4s ~2.4s
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4 Q8	128 tok/s 92 tok/s	~2.5s ~2.7s
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4 Q8	128 tok/s 92 tok/s	~2.5s ~2.7s
-	Hy-MT2 30B A3B	30B (3B active)	INT4 Q8	127 tok/s 91 tok/s	~2.5s ~2.8s
-	Spark X2.5 4B	4B	INT4 Q8 FP16	127 tok/s 88 tok/s 53 tok/s	~2.5s ~2.5s ~2.6s
-	Phi-2	2.7B	INT4 Q8 FP16	110 tok/s 89 tok/s 61 tok/s	~1.8s ~1.8s ~1.8s
-	MaLLaM-3B	3B	INT4 Q8 FP16	105 tok/s 84 tok/s 57 tok/s	~1.9s ~2.0s ~2.0s
-	SmolLM3 3B	3B	INT4 Q8 FP16	105 tok/s 84 tok/s 57 tok/s	~1.9s ~2.0s ~2.0s
-	Falcon-7B	7B	INT4 Q8 FP16	93 tok/s 59 tok/s 33 tok/s	~4.3s ~4.3s ~4.4s
-	Falcon3-7B	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	90 tok/s 58 tok/s 33 tok/s	~4.3s ~4.4s ~4.4s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	87 tok/s 56 tok/s 31 tok/s	~4.5s ~4.5s ~4.5s
-	LensVLM 9B	9B	INT4 Q8 FP16	75 tok/s 47 tok/s 26 tok/s	~5.6s ~5.6s ~5.6s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	75 tok/s 47 tok/s 26 tok/s	~5.6s ~5.6s ~5.6s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	73 tok/s 54 tok/s 33 tok/s	~3.7s ~3.8s ~3.8s
-	Yi-6B	6B	INT4 Q8 FP16	73 tok/s 54 tok/s 33 tok/s	~3.7s ~3.8s ~3.8s
-	Falcon3-10B	10B	INT4 Q8 FP16	69 tok/s 43 tok/s 24 tok/s	~6.2s ~6.2s ~6.2s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8 FP16	69 tok/s 61 tok/s 41 tok/s	~2.3s ~2.3s ~2.7s
-	Falcon2-11B	11B	INT4 Q8 FP16	67 tok/s 41 tok/s 22 tok/s	~6.8s ~6.8s ~6.8s
-	Hunyuan Lite	7B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.3s ~4.4s ~4.4s
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.3s ~4.4s ~4.4s
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.4s ~4.4s ~4.4s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	66 tok/s 48 tok/s 29 tok/s	~4.4s ~4.4s ~4.4s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	61 tok/s 43 tok/s 26 tok/s	~4.9s ~5.0s ~5.0s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	61 tok/s 43 tok/s 26 tok/s	~4.9s ~5.0s ~5.0s
-	GLM-4-9B	9B	INT4 Q8 FP16	56 tok/s 39 tok/s 23 tok/s	~5.6s ~5.6s ~5.6s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	56 tok/s 39 tok/s 23 tok/s	~5.6s ~5.6s ~5.6s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	56 tok/s 39 tok/s 23 tok/s	~5.6s ~5.6s ~5.6s
-	Yi-9B	9B	INT4 Q8 FP16	56 tok/s 39 tok/s 23 tok/s	~5.6s ~5.6s ~5.6s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	55 tok/s 39 tok/s 23 tok/s	~5.7s ~5.7s ~5.7s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	55 tok/s 39 tok/s 23 tok/s	~5.7s ~5.7s ~5.7s
-	GreenMind-14B-R1	14B	INT4 Q8 FP16	40 tok/s 27 tok/s 15 tok/s	~8.6s ~8.6s ~9.4s
-	Falcon-40B	40B	INT4	22 tok/s	~24.0s

Apple M5 Max (48GB)

VRAM Scaling by Context Length

Best Models to Run on Apple M5 Max (48GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions