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

Based on model rankings and 36 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~16 tok/s. Top Coding: Qwen3.5-27B (INT4) ~16 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~21 tok/s. Maximum Model Size: DeepSeek-R1 70B (INT4) ~6 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 Pro (48GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4 Q8	16 tok/s 9 tok/s	~32.6s ~35.5s
#48	Agnes 3.0 Flash	33B	INT4 Q8	13 tok/s 7 tok/s	~39.6s ~46.7s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4 Q8	67 tok/s 42 tok/s	~6.4s ~7.5s
#83	Qwen3.5-27B	27B	INT4 Q8	16 tok/s 9 tok/s	~32.6s ~35.5s
#92	Sarvam-30B	32B (2.4B active)	INT4 Q8	88 tok/s 62 tok/s	~4.4s ~4.8s
#94	Qwen3.6 35B A3B	35B (3B active)	INT4 Q8	79 tok/s 51 tok/s	~5.2s ~6.1s
#96	Muse Glimmer 30B	30B	INT4 Q8	15 tok/s 8 tok/s	~36.1s ~39.3s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8 FP16	21 tok/s 14 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8 FP16	2 tok/s 1 tok/s 1 tok/s	~356.5s ~357.0s ~421.2s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4 Q8	72 tok/s 49 tok/s	~5.8s ~5.8s
#109	Gemma 4 12B	11.95B	INT4 Q8 FP16	24 tok/s 16 tok/s 9 tok/s	~14.7s ~14.7s ~14.7s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	43 tok/s 26 tok/s 14 tok/s	~11.1s ~11.1s ~11.1s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4 Q8	79 tok/s 51 tok/s	~5.2s ~6.1s
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	14 tok/s 8 tok/s 4 tok/s	~35.0s ~35.0s ~35.1s
#129	Gemma 4 31B	30.7B	INT4 Q8	14 tok/s 8 tok/s	~36.9s ~40.2s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4 Q8	30 tok/s 25 tok/s	~5.1s ~6.0s
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	80 tok/s 52 tok/s 29 tok/s	~5.1s ~5.1s ~5.1s
#133	Qwen3-30B-A3B	30B (3B active)	INT4 Q8	80 tok/s 54 tok/s	~5.1s ~5.5s
#136	GPT-OSS 20B	21B (3.6B active)	INT4 Q8	34 tok/s 29 tok/s	~5.5s ~5.5s
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4 Q8	80 tok/s 54 tok/s	~5.1s ~5.5s
#153	Qwen3-32B	32B	INT4 Q8	14 tok/s 7 tok/s	~38.5s ~41.9s
#155	Mistral Large 3	41B (675B active)	INT4	1 tok/s	~845.9s
#156	GLM-4	32B	INT4	11 tok/s	~38.5s
#157	DeepSeek-R1 70B	70B	INT4	6 tok/s	~98.3s
#158	GLM-4V	9B	INT4 Q8 FP16	31 tok/s 21 tok/s 12 tok/s	~11.1s ~11.1s ~11.2s
#159	Gemma 3 27B	27B	INT4 Q8	16 tok/s 9 tok/s	~32.6s ~35.5s
#160	Phi-4	14B	INT4 Q8 FP16	29 tok/s 17 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#161	Llama 3.3 70B	70B	INT4	6 tok/s	~98.3s
#162	DeepSeek-R1 32B	32B	INT4 Q8	14 tok/s 7 tok/s	~38.5s ~41.9s
#163	Hunyuan Standard	52B (389B active)	INT4	1 tok/s	~528.9s
#164	Qwen2.5-32B	32B	INT4 Q8	14 tok/s 7 tok/s	~38.5s ~41.9s
#168	Magistral Small	24B	INT4 Q8	14 tok/s 8 tok/s	~29.1s ~31.6s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.7s ~8.7s ~8.7s
#170	OLMo 3.1 32B Instruct	32B	INT4	11 tok/s	~38.5s
#171	Llama 3.1 70B	70B	INT4	6 tok/s	~98.3s
#172	DeepSeek-R1 14B	14B	INT4 Q8 FP16	30 tok/s 17 tok/s 9 tok/s	~17.1s ~17.2s ~18.7s
#173	Ministral 3 14B	14B	INT4 Q8 FP16	21 tok/s 14 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~14.7s ~14.7s ~14.8s
#180	OLMo 3 32B Base	32B	INT4	11 tok/s	~38.5s
#181	Mistral-Small-2501	24B	INT4 Q8	18 tok/s 10 tok/s	~29.0s ~29.1s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	34 tok/s 23 tok/s 14 tok/s	~9.9s ~9.9s ~9.9s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	122 tok/s 86 tok/s 52 tok/s	~2.7s ~2.7s ~2.7s
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	122 tok/s 86 tok/s 52 tok/s	~2.7s ~2.7s ~2.7s
#185	Qwen2.5-14B	14B	INT4 Q8 FP16	29 tok/s 17 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#186	Gemma 2 27B	27B	INT4 Q8	16 tok/s 9 tok/s	~32.6s ~35.5s
#187	OLMo 3.1 32B Think	32B	INT4	11 tok/s	~38.5s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	67 tok/s 42 tok/s 23 tok/s	~6.4s ~6.4s ~6.4s
#189	Qwen3-4B	4B	INT4 Q8 FP16	80 tok/s 52 tok/s 29 tok/s	~5.1s ~5.1s ~5.1s
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	49 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#191	Llama 3 70B	70B	INT4	6 tok/s	~98.3s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	64 tok/s 49 tok/s 31 tok/s	~3.9s ~3.9s ~3.9s
#195	Qwen3-14B	14B	INT4 Q8 FP16	29 tok/s 17 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#196	Kimi Linear 48B A3B Instruct	48B (3B active)	INT4	24 tok/s	~6.0s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	43 tok/s 26 tok/s 14 tok/s	~11.1s ~11.1s ~11.1s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
#199	Qwen3-8B	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	80 tok/s 52 tok/s 29 tok/s	~5.1s ~5.1s ~5.1s
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	54 tok/s 33 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	134 tok/s 96 tok/s 59 tok/s	~2.3s ~2.3s ~2.3s
#205	Command R	35B	INT4	10 tok/s	~42.0s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	226 tok/s 206 tok/s 170 tok/s	~359 ms ~359 ms ~360 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	211 tok/s 173 tok/s 124 tok/s	~806 ms ~808 ms ~810 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	83 tok/s 54 tok/s 30 tok/s	~4.8s ~4.8s ~4.9s
#211	Mixtral-8x7B-v0.1	46.7B (7B active)	INT4	46 tok/s	~10.1s
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	47 tok/s 29 tok/s 15 tok/s	~9.9s ~9.9s ~9.9s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	143 tok/s 105 tok/s 65 tok/s	~2.0s ~2.0s ~2.0s
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	97 tok/s 65 tok/s 37 tok/s	~3.9s ~3.9s ~3.9s
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	190 tok/s 151 tok/s 103 tok/s	~1.1s ~1.1s ~1.1s
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	83 tok/s 54 tok/s 30 tok/s	~4.8s ~4.8s ~4.9s
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	64 tok/s 54 tok/s 38 tok/s	~2.9s ~2.9s ~2.9s
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	51 tok/s 31 tok/s 17 tok/s	~9.0s ~9.0s ~9.1s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	172 tok/s 133 tok/s 88 tok/s	~1.4s ~1.4s ~1.4s
#220	Gemma 2 2B	2B	INT4 Q8 FP16	122 tok/s 86 tok/s 52 tok/s	~2.7s ~2.7s ~2.7s
#221	Phi-3-medium	14B	INT4 Q8 FP16	29 tok/s 17 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.7s ~8.7s ~8.7s
#223	Yi-34B	34B	INT4	10 tok/s	~40.9s
#224	Qwen2-7B	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
#225	Phi-3-small	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	172 tok/s 133 tok/s 88 tok/s	~1.4s ~1.4s ~1.4s
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	51 tok/s 31 tok/s 17 tok/s	~9.0s ~9.0s ~9.1s
#229	Gemma 1 7B	7B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.7s ~8.7s ~8.7s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	131 tok/s 90 tok/s 53 tok/s	~2.7s ~2.7s ~2.7s
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	42 tok/s 30 tok/s 18 tok/s	~7.5s ~7.5s ~7.5s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	42 tok/s 30 tok/s 18 tok/s	~7.5s ~7.5s ~7.5s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	42 tok/s 30 tok/s 18 tok/s	~7.5s ~7.5s ~7.5s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	223 tok/s 187 tok/s 137 tok/s	~668 ms ~669 ms ~671 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	223 tok/s 187 tok/s 137 tok/s	~668 ms ~669 ms ~671 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	218 tok/s 198 tok/s 161 tok/s	~400 ms ~400 ms ~401 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	218 tok/s 198 tok/s 161 tok/s	~400 ms ~400 ms ~401 ms
-	Falcon-1B	1B	INT4 Q8 FP16	182 tok/s 138 tok/s 90 tok/s	~1.4s ~1.4s ~1.4s
-	Falcon3-1B	1B	INT4 Q8 FP16	172 tok/s 133 tok/s 88 tok/s	~1.4s ~1.4s ~1.4s
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	143 tok/s 105 tok/s 65 tok/s	~2.0s ~2.0s ~2.0s
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	143 tok/s 105 tok/s 65 tok/s	~2.0s ~2.0s ~2.0s
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	134 tok/s 96 tok/s 59 tok/s	~2.3s ~2.3s ~2.3s
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	130 tok/s 93 tok/s 57 tok/s	~2.4s ~2.4s ~2.4s
-	Falcon-3B	3B	INT4 Q8 FP16	103 tok/s 67 tok/s 38 tok/s	~3.9s ~3.9s ~3.9s
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	101 tok/s 84 tok/s 59 tok/s	~1.8s ~1.8s ~1.8s
-	Phi-1	1.3B	INT4 Q8 FP16	101 tok/s 84 tok/s 59 tok/s	~1.8s ~1.8s ~1.8s
-	Phi-1.5	1.3B	INT4 Q8 FP16	101 tok/s 84 tok/s 59 tok/s	~1.8s ~1.8s ~1.8s
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	98 tok/s 65 tok/s 37 tok/s	~3.9s ~3.9s ~3.9s
-	Falcon3-3B	3B	INT4 Q8 FP16	97 tok/s 65 tok/s 37 tok/s	~3.9s ~3.9s ~3.9s
-	Qwen2.5-3B	3B	INT4 Q8 FP16	97 tok/s 65 tok/s 37 tok/s	~3.9s ~3.9s ~3.9s
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	84 tok/s 59 tok/s	~4.8s ~4.8s
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	84 tok/s 59 tok/s	~4.8s ~4.8s
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4 Q8	81 tok/s 55 tok/s	~5.0s ~5.4s
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4 Q8	81 tok/s 55 tok/s	~5.0s ~5.4s
-	Hy-MT2 30B A3B	30B (3B active)	INT4 Q8	80 tok/s 54 tok/s	~5.1s ~5.5s
-	Spark X2.5 4B	4B	INT4 Q8 FP16	80 tok/s 52 tok/s 29 tok/s	~5.1s ~5.1s ~5.1s
-	Phi-2	2.7B	INT4 Q8 FP16	67 tok/s 52 tok/s 34 tok/s	~3.5s ~3.5s ~3.5s
-	MaLLaM-3B	3B	INT4 Q8 FP16	64 tok/s 49 tok/s 31 tok/s	~3.9s ~3.9s ~3.9s
-	SmolLM3 3B	3B	INT4 Q8 FP16	64 tok/s 49 tok/s 31 tok/s	~3.9s ~3.9s ~3.9s
-	Falcon-7B	7B	INT4 Q8 FP16	55 tok/s 33 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
-	Falcon3-7B	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	53 tok/s 32 tok/s 17 tok/s	~8.7s ~8.7s ~8.7s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	51 tok/s 31 tok/s 17 tok/s	~9.0s ~9.0s ~9.1s
-	LensVLM 9B	9B	INT4 Q8 FP16	43 tok/s 26 tok/s 14 tok/s	~11.1s ~11.1s ~11.1s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	43 tok/s 26 tok/s 14 tok/s	~11.1s ~11.1s ~11.1s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	42 tok/s 30 tok/s 18 tok/s	~7.5s ~7.5s ~7.5s
-	Yi-6B	6B	INT4 Q8 FP16	42 tok/s 30 tok/s 18 tok/s	~7.5s ~7.5s ~7.5s
-	Falcon3-10B	10B	INT4 Q8 FP16	39 tok/s 23 tok/s 12 tok/s	~12.3s ~12.3s ~12.3s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8 FP16	39 tok/s 34 tok/s 22 tok/s	~4.6s ~4.6s ~5.4s
-	Falcon2-11B	11B	INT4 Q8 FP16	38 tok/s 22 tok/s 11 tok/s	~13.5s ~13.5s ~13.6s
-	Hunyuan Lite	7B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.7s ~8.7s ~8.7s
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.7s ~8.7s ~8.7s
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.8s ~8.8s ~8.8s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	37 tok/s 26 tok/s 15 tok/s	~8.8s ~8.8s ~8.8s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	34 tok/s 23 tok/s 14 tok/s	~9.9s ~9.9s ~9.9s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	34 tok/s 23 tok/s 14 tok/s	~9.9s ~9.9s ~9.9s
-	GLM-4-9B	9B	INT4 Q8 FP16	31 tok/s 21 tok/s 12 tok/s	~11.1s ~11.1s ~11.2s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	31 tok/s 21 tok/s 12 tok/s	~11.1s ~11.1s ~11.2s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	31 tok/s 21 tok/s 12 tok/s	~11.1s ~11.1s ~11.2s
-	Yi-9B	9B	INT4 Q8 FP16	31 tok/s 21 tok/s 12 tok/s	~11.1s ~11.1s ~11.2s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	30 tok/s 21 tok/s 12 tok/s	~11.3s ~11.4s ~11.4s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	30 tok/s 21 tok/s 12 tok/s	~11.3s ~11.4s ~11.4s
-	GreenMind-14B-R1	14B	INT4 Q8 FP16	21 tok/s 14 tok/s 8 tok/s	~17.1s ~17.2s ~18.7s
-	Falcon-40B	40B	INT4	11 tok/s	~47.9s

Apple M5 Pro (48GB)

VRAM Scaling by Context Length

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

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions