What are the best models to run on NVIDIA RTX A5000 (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~38 tok/s. Top Coding: Qwen3.5-27B (INT4) ~38 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~49 tok/s. Maximum Model Size: Falcon-40B (INT4) ~26 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 NVIDIA RTX A5000 (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	38 tok/s	~2.7s
#48	Agnes 3.0 Flash	33B	INT4	32 tok/s	~3.2s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	130 tok/s	~569 ms
#83	Qwen3.5-27B	27B	INT4	38 tok/s	~2.7s
#92	Sarvam-30B	32B (2.4B active)	INT4	169 tok/s	~367 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	147 tok/s	~464 ms
#96	Muse Glimmer 30B	30B	INT4	35 tok/s	~2.9s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	49 tok/s 34 tok/s	~1.4s ~1.4s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	4 tok/s 2 tok/s	~29.0s ~31.7s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	145 tok/s	~480 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	56 tok/s 39 tok/s	~1.2s ~1.2s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	93 tok/s 59 tok/s 31 tok/s	~908 ms ~915 ms ~1.0s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	147 tok/s	~464 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	33 tok/s 20 tok/s 10 tok/s	~2.9s ~2.9s ~2.9s
#129	Gemma 4 31B	30.7B	INT4	34 tok/s	~3.0s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	65 tok/s	~464 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	157 tok/s 109 tok/s 65 tok/s	~420 ms ~423 ms ~429 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	157 tok/s	~419 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	76 tok/s	~456 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	157 tok/s	~419 ms
#153	Qwen3-32B	32B	INT4	33 tok/s	~3.1s
#156	GLM-4	32B	INT4	23 tok/s	~3.7s
#158	GLM-4V	9B	INT4 Q8 FP16	69 tok/s 49 tok/s 28 tok/s	~912 ms ~918 ms ~1.0s
#159	Gemma 3 27B	27B	INT4	38 tok/s	~2.7s
#160	Phi-4	14B	INT4 Q8	66 tok/s 40 tok/s	~1.4s ~1.4s
#162	DeepSeek-R1 32B	32B	INT4	34 tok/s	~3.1s
#164	Qwen2.5-32B	32B	INT4	33 tok/s	~3.1s
#168	Magistral Small	24B	INT4	32 tok/s	~2.4s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	83 tok/s 60 tok/s 36 tok/s	~715 ms ~720 ms ~731 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	23 tok/s	~3.7s
#172	DeepSeek-R1 14B	14B	INT4 Q8	68 tok/s 41 tok/s	~1.4s ~1.4s
#173	Ministral 3 14B	14B	INT4 Q8	49 tok/s 34 tok/s	~1.4s ~1.4s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	75 tok/s 46 tok/s 22 tok/s	~1.2s ~1.2s ~1.4s
#180	OLMo 3 32B Base	32B	INT4	23 tok/s	~3.7s
#181	Mistral-Small-2501	24B	INT4	42 tok/s	~2.4s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	76 tok/s 54 tok/s 31 tok/s	~813 ms ~818 ms ~901 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~226 ms ~227 ms ~230 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~226 ms ~227 ms ~230 ms
#185	Qwen2.5-14B	14B	INT4 Q8	66 tok/s 40 tok/s	~1.4s ~1.4s
#186	Gemma 2 27B	27B	INT4	38 tok/s	~2.7s
#187	OLMo 3.1 32B Think	32B	INT4	23 tok/s	~3.7s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	136 tok/s 92 tok/s 54 tok/s	~527 ms ~531 ms ~538 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	157 tok/s 109 tok/s 65 tok/s	~420 ms ~423 ms ~429 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	104 tok/s 66 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	131 tok/s 105 tok/s 71 tok/s	~325 ms ~327 ms ~331 ms
#195	Qwen3-14B	14B	INT4 Q8	66 tok/s 40 tok/s	~1.4s ~1.4s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	93 tok/s 59 tok/s 31 tok/s	~908 ms ~915 ms ~1.0s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	157 tok/s 109 tok/s 65 tok/s	~420 ms ~423 ms ~429 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	114 tok/s 73 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	229 tok/s 181 tok/s 123 tok/s	~194 ms ~195 ms ~198 ms
#205	Command R	35B	INT4	22 tok/s	~4.0s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	317 tok/s 301 tok/s 268 tok/s	~36 ms ~36 ms ~37 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	305 tok/s 271 tok/s 217 tok/s	~73 ms ~73 ms ~74 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~400 ms ~403 ms ~409 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	102 tok/s 65 tok/s 36 tok/s	~809 ms ~815 ms ~827 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	240 tok/s 192 tok/s 133 tok/s	~173 ms ~174 ms ~176 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	182 tok/s 132 tok/s 82 tok/s	~323 ms ~325 ms ~329 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	287 tok/s 248 tok/s 190 tok/s	~96 ms ~96 ms ~98 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	161 tok/s 113 tok/s 68 tok/s	~400 ms ~403 ms ~409 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	131 tok/s 114 tok/s 84 tok/s	~244 ms ~245 ms ~248 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	108 tok/s 70 tok/s 39 tok/s	~741 ms ~746 ms ~757 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	270 tok/s 228 tok/s 169 tok/s	~120 ms ~121 ms ~122 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	215 tok/s 166 tok/s 110 tok/s	~226 ms ~227 ms ~230 ms
#221	Phi-3-medium	14B	INT4 Q8	66 tok/s 40 tok/s	~1.4s ~1.4s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	83 tok/s 60 tok/s 36 tok/s	~715 ms ~720 ms ~731 ms
#223	Yi-34B	34B	INT4	22 tok/s	~3.9s
#224	Qwen2-7B	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	270 tok/s 228 tok/s 169 tok/s	~120 ms ~121 ms ~122 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	108 tok/s 70 tok/s 39 tok/s	~741 ms ~746 ms ~757 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	83 tok/s 60 tok/s 36 tok/s	~715 ms ~720 ms ~731 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	226 tok/s 172 tok/s 112 tok/s	~226 ms ~227 ms ~230 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	91 tok/s 67 tok/s 41 tok/s	~618 ms ~622 ms ~631 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	91 tok/s 67 tok/s 41 tok/s	~618 ms ~622 ms ~631 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	91 tok/s 67 tok/s 41 tok/s	~618 ms ~622 ms ~631 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	315 tok/s 284 tok/s 233 tok/s	~61 ms ~62 ms ~62 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	315 tok/s 284 tok/s 233 tok/s	~61 ms ~62 ms ~62 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	311 tok/s 294 tok/s 259 tok/s	~40 ms ~40 ms ~40 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	311 tok/s 294 tok/s 259 tok/s	~40 ms ~40 ms ~40 ms
-	Falcon-1B	1B	INT4 Q8 FP16	280 tok/s 234 tok/s 172 tok/s	~120 ms ~121 ms ~122 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	270 tok/s 228 tok/s 169 tok/s	~120 ms ~121 ms ~122 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	240 tok/s 192 tok/s 133 tok/s	~173 ms ~174 ms ~176 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	240 tok/s 192 tok/s 133 tok/s	~173 ms ~174 ms ~176 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	229 tok/s 181 tok/s 123 tok/s	~194 ms ~195 ms ~198 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	225 tok/s 176 tok/s 118 tok/s	~204 ms ~205 ms ~208 ms
-	Falcon-3B	3B	INT4 Q8 FP16	190 tok/s 136 tok/s 84 tok/s	~322 ms ~324 ms ~329 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	187 tok/s 163 tok/s 123 tok/s	~152 ms ~153 ms ~155 ms
-	Phi-1	1.3B	INT4 Q8 FP16	187 tok/s 163 tok/s 123 tok/s	~152 ms ~153 ms ~155 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	187 tok/s 163 tok/s 123 tok/s	~152 ms ~153 ms ~155 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	184 tok/s 133 tok/s 82 tok/s	~322 ms ~325 ms ~329 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	182 tok/s 132 tok/s 82 tok/s	~323 ms ~325 ms ~329 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	182 tok/s 132 tok/s 82 tok/s	~323 ms ~325 ms ~329 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	163 tok/s 110 tok/s	~393 ms ~463 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	163 tok/s 110 tok/s	~393 ms ~463 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	158 tok/s	~413 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	158 tok/s	~413 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	157 tok/s	~419 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	157 tok/s 109 tok/s 65 tok/s	~420 ms ~423 ms ~429 ms
-	Phi-2	2.7B	INT4 Q8 FP16	136 tok/s 111 tok/s 76 tok/s	~296 ms ~297 ms ~301 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	131 tok/s 105 tok/s 71 tok/s	~325 ms ~327 ms ~331 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	131 tok/s 105 tok/s 71 tok/s	~325 ms ~327 ms ~331 ms
-	Falcon-7B	7B	INT4 Q8 FP16	116 tok/s 74 tok/s 41 tok/s	~712 ms ~717 ms ~727 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	111 tok/s 72 tok/s 41 tok/s	~712 ms ~717 ms ~728 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	108 tok/s 70 tok/s 39 tok/s	~741 ms ~746 ms ~757 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	93 tok/s 59 tok/s 31 tok/s	~908 ms ~915 ms ~1.0s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	93 tok/s 59 tok/s 31 tok/s	~908 ms ~915 ms ~1.0s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	91 tok/s 67 tok/s 41 tok/s	~618 ms ~622 ms ~631 ms
-	Yi-6B	6B	INT4 Q8 FP16	91 tok/s 67 tok/s 41 tok/s	~618 ms ~622 ms ~631 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	86 tok/s 54 tok/s 28 tok/s	~1.0s ~1.0s ~1.1s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	86 tok/s 72 tok/s	~383 ms ~417 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	83 tok/s 51 tok/s 25 tok/s	~1.1s ~1.1s ~1.3s
-	Hunyuan Lite	7B	INT4 Q8 FP16	83 tok/s 60 tok/s 36 tok/s	~715 ms ~720 ms ~731 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	83 tok/s 60 tok/s 36 tok/s	~715 ms ~720 ms ~731 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	82 tok/s 59 tok/s 36 tok/s	~725 ms ~730 ms ~740 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	82 tok/s 59 tok/s 36 tok/s	~725 ms ~730 ms ~740 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	76 tok/s 54 tok/s 31 tok/s	~813 ms ~818 ms ~901 ms
-	Typhoon-2-8B	8B	INT4 Q8 FP16	76 tok/s 54 tok/s 31 tok/s	~813 ms ~818 ms ~901 ms
-	GLM-4-9B	9B	INT4 Q8 FP16	69 tok/s 49 tok/s 28 tok/s	~912 ms ~918 ms ~1.0s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	69 tok/s 49 tok/s 28 tok/s	~912 ms ~918 ms ~1.0s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	69 tok/s 49 tok/s 28 tok/s	~912 ms ~918 ms ~1.0s
-	Yi-9B	9B	INT4 Q8 FP16	69 tok/s 49 tok/s 28 tok/s	~912 ms ~918 ms ~1.0s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	68 tok/s 48 tok/s 27 tok/s	~931 ms ~937 ms ~1.0s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	68 tok/s 48 tok/s 27 tok/s	~931 ms ~937 ms ~1.0s
-	GreenMind-14B-R1	14B	INT4 Q8	49 tok/s 34 tok/s	~1.4s ~1.4s
-	Falcon-40B	40B	INT4	26 tok/s	~4.2s

NVIDIA RTX A5000 (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX A5000 (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions