What are the best models to run on NVIDIA A10G (24GB)?

Based on model rankings and 24 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) ~39 tok/s. Maximum Model Size: Falcon-40B (INT4) ~20 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 A10G (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	30 tok/s	~4.8s
#48	Agnes 3.0 Flash	33B	INT4	25 tok/s	~5.8s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	122 tok/s	~1.0s
#83	Qwen3.5-27B	27B	INT4	30 tok/s	~4.8s
#92	Sarvam-30B	32B (2.4B active)	INT4	171 tok/s	~652 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	144 tok/s	~827 ms
#96	Muse Glimmer 30B	30B	INT4	27 tok/s	~5.3s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	39 tok/s 26 tok/s	~2.5s ~2.5s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	3 tok/s 1 tok/s	~52.2s ~57.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	138 tok/s	~855 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	45 tok/s 30 tok/s	~2.2s ~2.2s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	81 tok/s 48 tok/s 24 tok/s	~1.6s ~1.6s ~1.8s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	144 tok/s	~827 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	26 tok/s 15 tok/s 8 tok/s	~5.1s ~5.2s ~5.2s
#129	Gemma 4 31B	30.7B	INT4	26 tok/s	~5.4s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	54 tok/s	~822 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	154 tok/s 97 tok/s 54 tok/s	~746 ms ~750 ms ~758 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	154 tok/s	~746 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	63 tok/s	~808 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	154 tok/s	~746 ms
#153	Qwen3-32B	32B	INT4	25 tok/s	~5.6s
#156	GLM-4	32B	INT4	18 tok/s	~6.6s
#158	GLM-4V	9B	INT4 Q8 FP16	57 tok/s 39 tok/s 21 tok/s	~1.6s ~1.6s ~1.8s
#159	Gemma 3 27B	27B	INT4	30 tok/s	~4.8s
#160	Phi-4	14B	INT4 Q8	54 tok/s 32 tok/s	~2.5s ~2.5s
#162	DeepSeek-R1 32B	32B	INT4	26 tok/s	~5.6s
#164	Qwen2.5-32B	32B	INT4	25 tok/s	~5.6s
#168	Magistral Small	24B	INT4	25 tok/s	~4.3s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	70 tok/s 49 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
#170	OLMo 3.1 32B Instruct	32B	INT4	18 tok/s	~6.6s
#172	DeepSeek-R1 14B	14B	INT4 Q8	56 tok/s 32 tok/s	~2.5s ~2.5s
#173	Ministral 3 14B	14B	INT4 Q8	39 tok/s 26 tok/s	~2.5s ~2.5s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	63 tok/s 37 tok/s 17 tok/s	~2.2s ~2.2s ~2.6s
#180	OLMo 3 32B Base	32B	INT4	18 tok/s	~6.6s
#181	Mistral-Small-2501	24B	INT4	33 tok/s	~4.3s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	63 tok/s 44 tok/s 24 tok/s	~1.5s ~1.5s ~1.6s
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	241 tok/s 166 tok/s 98 tok/s	~397 ms ~399 ms ~403 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	241 tok/s 166 tok/s 98 tok/s	~397 ms ~399 ms ~403 ms
#185	Qwen2.5-14B	14B	INT4 Q8	54 tok/s 32 tok/s	~2.5s ~2.5s
#186	Gemma 2 27B	27B	INT4	30 tok/s	~4.8s
#187	OLMo 3.1 32B Think	32B	INT4	18 tok/s	~6.6s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	128 tok/s 79 tok/s 43 tok/s	~939 ms ~944 ms ~955 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	154 tok/s 97 tok/s 54 tok/s	~746 ms ~750 ms ~758 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	92 tok/s 54 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#193	Ministral 3 3B	3B	INT4 Q8 FP16	121 tok/s 92 tok/s 59 tok/s	~574 ms ~576 ms ~583 ms
#195	Qwen3-14B	14B	INT4 Q8	54 tok/s 32 tok/s	~2.5s ~2.5s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	81 tok/s 48 tok/s 24 tok/s	~1.6s ~1.6s ~1.8s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
#199	Qwen3-8B	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#200	Gemma 3 4B	4B	INT4 Q8 FP16	154 tok/s 97 tok/s 54 tok/s	~746 ms ~750 ms ~758 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	102 tok/s 61 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	266 tok/s 187 tok/s 112 tok/s	~340 ms ~341 ms ~345 ms
#205	Command R	35B	INT4	16 tok/s	~7.2s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	473 tok/s 426 tok/s 344 tok/s	~55 ms ~55 ms ~56 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	437 tok/s 351 tok/s 244 tok/s	~120 ms ~121 ms ~122 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	160 tok/s 102 tok/s 56 tok/s	~711 ms ~715 ms ~723 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	89 tok/s 53 tok/s 28 tok/s	~1.4s ~1.5s ~1.5s
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	286 tok/s 204 tok/s 125 tok/s	~301 ms ~302 ms ~305 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	188 tok/s 123 tok/s 70 tok/s	~571 ms ~574 ms ~580 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	390 tok/s 302 tok/s 201 tok/s	~162 ms ~163 ms ~165 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	160 tok/s 102 tok/s 56 tok/s	~711 ms ~715 ms ~723 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	121 tok/s 102 tok/s 72 tok/s	~428 ms ~429 ms ~433 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	96 tok/s 58 tok/s 31 tok/s	~1.3s ~1.3s ~1.3s
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	350 tok/s 264 tok/s 170 tok/s	~206 ms ~207 ms ~209 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	241 tok/s 166 tok/s 98 tok/s	~397 ms ~399 ms ~403 ms
#221	Phi-3-medium	14B	INT4 Q8	54 tok/s 32 tok/s	~2.5s ~2.5s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	70 tok/s 49 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
#223	Yi-34B	34B	INT4	17 tok/s	~7.0s
#224	Qwen2-7B	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
#225	Phi-3-small	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
#226	Gemma 3 1B	1B	INT4 Q8 FP16	350 tok/s 264 tok/s 170 tok/s	~206 ms ~207 ms ~209 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	96 tok/s 58 tok/s 31 tok/s	~1.3s ~1.3s ~1.3s
#229	Gemma 1 7B	7B	INT4 Q8 FP16	70 tok/s 49 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
#230	Gemma 1 2B	2B	INT4 Q8 FP16	259 tok/s 174 tok/s 101 tok/s	~397 ms ~398 ms ~403 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	78 tok/s 55 tok/s 33 tok/s	~1.1s ~1.1s ~1.1s
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	78 tok/s 55 tok/s 33 tok/s	~1.1s ~1.1s ~1.1s
#233	ChatGLM-6B	6B	INT4 Q8 FP16	78 tok/s 55 tok/s 33 tok/s	~1.1s ~1.1s ~1.1s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	465 tok/s 382 tok/s 273 tok/s	~100 ms ~101 ms ~102 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	465 tok/s 382 tok/s 273 tok/s	~100 ms ~101 ms ~102 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	454 tok/s 407 tok/s 324 tok/s	~61 ms ~61 ms ~62 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	454 tok/s 407 tok/s 324 tok/s	~61 ms ~61 ms ~62 ms
-	Falcon-1B	1B	INT4 Q8 FP16	372 tok/s 275 tok/s 175 tok/s	~206 ms ~207 ms ~209 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	350 tok/s 264 tok/s 170 tok/s	~206 ms ~207 ms ~209 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	286 tok/s 204 tok/s 125 tok/s	~301 ms ~302 ms ~305 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	286 tok/s 204 tok/s 125 tok/s	~301 ms ~302 ms ~305 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	266 tok/s 187 tok/s 112 tok/s	~340 ms ~341 ms ~345 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	258 tok/s 180 tok/s 107 tok/s	~358 ms ~359 ms ~363 ms
-	Falcon-3B	3B	INT4 Q8 FP16	200 tok/s 128 tok/s 71 tok/s	~571 ms ~573 ms ~580 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	196 tok/s 162 tok/s 113 tok/s	~264 ms ~265 ms ~268 ms
-	Phi-1	1.3B	INT4 Q8 FP16	196 tok/s 162 tok/s 113 tok/s	~264 ms ~265 ms ~268 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	196 tok/s 162 tok/s 113 tok/s	~264 ms ~265 ms ~268 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	191 tok/s 124 tok/s 70 tok/s	~571 ms ~574 ms ~580 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	188 tok/s 123 tok/s 70 tok/s	~571 ms ~574 ms ~580 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	188 tok/s 123 tok/s 70 tok/s	~571 ms ~574 ms ~580 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	162 tok/s 101 tok/s	~698 ms ~824 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	162 tok/s 101 tok/s	~698 ms ~824 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	156 tok/s	~734 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	156 tok/s	~734 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	154 tok/s	~746 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	154 tok/s 97 tok/s 54 tok/s	~746 ms ~750 ms ~758 ms
-	Phi-2	2.7B	INT4 Q8 FP16	128 tok/s 99 tok/s 64 tok/s	~521 ms ~524 ms ~529 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	121 tok/s 92 tok/s 59 tok/s	~574 ms ~576 ms ~583 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	121 tok/s 92 tok/s 59 tok/s	~574 ms ~576 ms ~583 ms
-	Falcon-7B	7B	INT4 Q8 FP16	104 tok/s 62 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
-	Falcon3-7B	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
-	Hy-MT2 7B	7B	INT4 Q8 FP16	100 tok/s 60 tok/s 32 tok/s	~1.3s ~1.3s ~1.3s
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	96 tok/s 58 tok/s 31 tok/s	~1.3s ~1.3s ~1.3s
-	LensVLM 9B	9B	INT4 Q8 FP16	81 tok/s 48 tok/s 24 tok/s	~1.6s ~1.6s ~1.8s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	81 tok/s 48 tok/s 24 tok/s	~1.6s ~1.6s ~1.8s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	78 tok/s 55 tok/s 33 tok/s	~1.1s ~1.1s ~1.1s
-	Yi-6B	6B	INT4 Q8 FP16	78 tok/s 55 tok/s 33 tok/s	~1.1s ~1.1s ~1.1s
-	Falcon3-10B	10B	INT4 Q8 FP16	74 tok/s 43 tok/s 22 tok/s	~1.8s ~1.8s ~2.0s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	74 tok/s 60 tok/s	~677 ms ~738 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	71 tok/s 41 tok/s 19 tok/s	~2.0s ~2.0s ~2.4s
-	Hunyuan Lite	7B	INT4 Q8 FP16	70 tok/s 49 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	70 tok/s 49 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	69 tok/s 48 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	69 tok/s 48 tok/s 28 tok/s	~1.3s ~1.3s ~1.3s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	63 tok/s 44 tok/s 24 tok/s	~1.5s ~1.5s ~1.6s
-	Typhoon-2-8B	8B	INT4 Q8 FP16	63 tok/s 44 tok/s 24 tok/s	~1.5s ~1.5s ~1.6s
-	GLM-4-9B	9B	INT4 Q8 FP16	57 tok/s 39 tok/s 21 tok/s	~1.6s ~1.6s ~1.8s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	57 tok/s 39 tok/s 21 tok/s	~1.6s ~1.6s ~1.8s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	57 tok/s 39 tok/s 21 tok/s	~1.6s ~1.6s ~1.8s
-	Yi-9B	9B	INT4 Q8 FP16	57 tok/s 39 tok/s 21 tok/s	~1.6s ~1.6s ~1.8s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	56 tok/s 38 tok/s 21 tok/s	~1.7s ~1.7s ~1.8s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	56 tok/s 38 tok/s 21 tok/s	~1.7s ~1.7s ~1.8s
-	GreenMind-14B-R1	14B	INT4 Q8	39 tok/s 26 tok/s	~2.5s ~2.5s
-	Falcon-40B	40B	INT4	20 tok/s	~7.6s

NVIDIA A10G (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA A10G (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions