What are the best models to run on NVIDIA L4 (24GB)?

Based on model rankings and 24 GB usable VRAM: Top Overall: Qwen 3.8 27B (INT4) ~15 tok/s. Top Coding: Qwen3.5-27B (INT4) ~15 tok/s. Optimal Daily Driver: Phi-4 Reasoning Plus (INT4) ~20 tok/s. Maximum Model Size: Falcon-40B (INT4) ~10 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 L4 (24GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#33	Qwen 3.8 27B	27B	INT4	15 tok/s	~2.8s
#48	Agnes 3.0 Flash	33B	INT4	13 tok/s	~3.4s
#72	K2 Horizon MoVA 36B A4B	36B (4B active)	INT4	68 tok/s	~599 ms
#83	Qwen3.5-27B	27B	INT4	15 tok/s	~2.8s
#92	Sarvam-30B	32B (2.4B active)	INT4	98 tok/s	~385 ms
#94	Qwen3.6 35B A3B	35B (3B active)	INT4	81 tok/s	~487 ms
#96	Muse Glimmer 30B	30B	INT4	14 tok/s	~3.1s
#100	Phi-4 Reasoning Plus	14B	INT4 Q8	20 tok/s 13 tok/s	~1.5s ~1.5s
#101	MiMo V2 Flash	15B (309B active)	INT4 Q8	1 tok/s 1 tok/s	~30.7s ~34.0s
#108	Gemma 4 26B A4B	25.2B (3.8B active)	INT4	77 tok/s	~505 ms
#109	Gemma 4 12B	11.95B	INT4 Q8	23 tok/s 16 tok/s	~1.3s ~1.3s
#113	Qwen3.5-9B	9B	INT4 Q8 FP16	43 tok/s 25 tok/s 12 tok/s	~958 ms ~975 ms ~1.1s
#114	Qwen3.5-35B-A3B	35B (3B active)	INT4	81 tok/s	~487 ms
#118	K2 Horizon 7B	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
#128	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8 FP16	13 tok/s 8 tok/s 4 tok/s	~3.0s ~3.1s ~3.2s
#129	Gemma 4 31B	30.7B	INT4	13 tok/s	~3.2s
#131	Nemotron 3.5 Lightning	30B (3B active)	INT4	28 tok/s	~501 ms
#132	Qwen3.5-4B	4B	INT4 Q8 FP16	87 tok/s 53 tok/s 28 tok/s	~441 ms ~448 ms ~465 ms
#133	Qwen3-30B-A3B	30B (3B active)	INT4	87 tok/s	~440 ms
#136	GPT-OSS 20B	21B (3.6B active)	INT4	33 tok/s	~489 ms
#146	Granite 4.2 8B	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#150	Gemma 4 E4B	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#151	Nemotron 3.5 Lightning 30B A3B	30B (3B active)	INT4	87 tok/s	~440 ms
#153	Qwen3-32B	32B	INT4	13 tok/s	~3.3s
#156	GLM-4	32B	INT4	9 tok/s	~3.9s
#158	GLM-4V	9B	INT4 Q8 FP16	30 tok/s 20 tok/s 11 tok/s	~968 ms ~984 ms ~1.1s
#159	Gemma 3 27B	27B	INT4	15 tok/s	~2.8s
#160	Phi-4	14B	INT4 Q8	28 tok/s 16 tok/s	~1.5s ~1.5s
#162	DeepSeek-R1 32B	32B	INT4	13 tok/s	~3.3s
#164	Qwen2.5-32B	32B	INT4	13 tok/s	~3.3s
#168	Magistral Small	24B	INT4	13 tok/s	~2.5s
#169	OLMo 3 7B Think	7B	INT4 Q8 FP16	37 tok/s 25 tok/s 15 tok/s	~759 ms ~771 ms ~800 ms
#170	OLMo 3.1 32B Instruct	32B	INT4	9 tok/s	~3.9s
#172	DeepSeek-R1 14B	14B	INT4 Q8	29 tok/s 16 tok/s	~1.5s ~1.5s
#173	Ministral 3 14B	14B	INT4 Q8	20 tok/s 13 tok/s	~1.5s ~1.5s
#177	Gemma 3 12B	12B	INT4 Q8 FP16	33 tok/s 19 tok/s 9 tok/s	~1.3s ~1.3s ~1.6s
#180	OLMo 3 32B Base	32B	INT4	9 tok/s	~3.9s
#181	Mistral-Small-2501	24B	INT4	17 tok/s	~2.5s
#182	Ministral 3 8B	8B	INT4 Q8 FP16	33 tok/s 22 tok/s 12 tok/s	~862 ms ~877 ms ~986 ms
#183	Qwen3.5-2B	2B	INT4 Q8 FP16	147 tok/s 95 tok/s 53 tok/s	~235 ms ~239 ms ~247 ms
#184	MiniCPM5-2B	2B	INT4 Q8 FP16	147 tok/s 95 tok/s 53 tok/s	~235 ms ~239 ms ~247 ms
#185	Qwen2.5-14B	14B	INT4 Q8	28 tok/s 16 tok/s	~1.5s ~1.5s
#186	Gemma 2 27B	27B	INT4	15 tok/s	~2.8s
#187	OLMo 3.1 32B Think	32B	INT4	9 tok/s	~3.9s
#188	Gemma 4 E2B	5.1B	INT4 Q8 FP16	71 tok/s 42 tok/s 22 tok/s	~554 ms ~564 ms ~585 ms
#189	Qwen3-4B	4B	INT4 Q8 FP16	87 tok/s 53 tok/s 28 tok/s	~441 ms ~448 ms ~465 ms
#190	DeepSeek-R1 8B	8B	INT4 Q8 FP16	49 tok/s 28 tok/s 15 tok/s	~853 ms ~868 ms ~901 ms
#193	Ministral 3 3B	3B	INT4 Q8 FP16	67 tok/s 50 tok/s 31 tok/s	~343 ms ~348 ms ~361 ms
#195	Qwen3-14B	14B	INT4 Q8	28 tok/s 16 tok/s	~1.5s ~1.5s
#197	Gemma 2 9B	9B	INT4 Q8 FP16	43 tok/s 25 tok/s 12 tok/s	~958 ms ~975 ms ~1.1s
#198	Qwen2.5-7B	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
#199	Qwen3-8B	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#200	Gemma 3 4B	4B	INT4 Q8 FP16	87 tok/s 53 tok/s 28 tok/s	~441 ms ~448 ms ~465 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8 FP16	55 tok/s 32 tok/s 17 tok/s	~750 ms ~763 ms ~792 ms
#202	Llama 3.1 8B	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#204	Qwen3-1.7B	1.7B	INT4 Q8 FP16	166 tok/s 109 tok/s 61 tok/s	~201 ms ~204 ms ~211 ms
#205	Command R	35B	INT4	8 tok/s	~4.3s
#206	Gemma 3 270M	0.27B	INT4 Q8 FP16	366 tok/s 313 tok/s 232 tok/s	~34 ms ~35 ms ~36 ms
#208	Llama 3 8B	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#209	Qwen3-0.6B	0.6B	INT4 Q8 FP16	326 tok/s 238 tok/s 149 tok/s	~72 ms ~73 ms ~76 ms
#210	Phi-4-Mini	3.8B	INT4 Q8 FP16	91 tok/s 55 tok/s 29 tok/s	~420 ms ~427 ms ~443 ms
#212	Ministral-8B-2410	8B	INT4 Q8 FP16	48 tok/s 28 tok/s 14 tok/s	~853 ms ~868 ms ~902 ms
#213	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	182 tok/s 120 tok/s 69 tok/s	~178 ms ~181 ms ~187 ms
#214	Llama 3.2 3B	3B	INT4 Q8 FP16	110 tok/s 68 tok/s 37 tok/s	~337 ms ~343 ms ~355 ms
#215	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	275 tok/s 195 tok/s 118 tok/s	~97 ms ~98 ms ~102 ms
#216	Phi-3-mini	3.8B	INT4 Q8 FP16	91 tok/s 55 tok/s 29 tok/s	~420 ms ~427 ms ~443 ms
#217	Gemma 3n E2B IT	6B (2B active)	INT4 Q8 FP16	67 tok/s 55 tok/s 38 tok/s	~259 ms ~262 ms ~270 ms
#218	Mistral-7B-Instruct-v0.1	7.3B	INT4 Q8 FP16	52 tok/s 30 tok/s 16 tok/s	~781 ms ~794 ms ~825 ms
#219	Llama 3.2 1B	1B	INT4 Q8 FP16	237 tok/s 164 tok/s 98 tok/s	~122 ms ~124 ms ~129 ms
#220	Gemma 2 2B	2B	INT4 Q8 FP16	147 tok/s 95 tok/s 53 tok/s	~235 ms ~239 ms ~247 ms
#221	Phi-3-medium	14B	INT4 Q8	28 tok/s 16 tok/s	~1.5s ~1.5s
#222	OLMo 3 7B Instruct	7B	INT4 Q8 FP16	37 tok/s 25 tok/s 15 tok/s	~759 ms ~771 ms ~800 ms
#223	Yi-34B	34B	INT4	9 tok/s	~4.2s
#224	Qwen2-7B	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
#225	Phi-3-small	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
#226	Gemma 3 1B	1B	INT4 Q8 FP16	237 tok/s 164 tok/s 98 tok/s	~122 ms ~124 ms ~129 ms
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4 Q8 FP16	52 tok/s 30 tok/s 16 tok/s	~781 ms ~794 ms ~825 ms
#229	Gemma 1 7B	7B	INT4 Q8 FP16	37 tok/s 25 tok/s 15 tok/s	~759 ms ~771 ms ~800 ms
#230	Gemma 1 2B	2B	INT4 Q8 FP16	161 tok/s 100 tok/s 55 tok/s	~234 ms ~238 ms ~246 ms
#231	ChatGLM3-6B	6B	INT4 Q8 FP16	42 tok/s 29 tok/s 17 tok/s	~655 ms ~665 ms ~690 ms
#232	ChatGLM2-6B	6B	INT4 Q8 FP16	42 tok/s 29 tok/s 17 tok/s	~655 ms ~665 ms ~690 ms
#233	ChatGLM-6B	6B	INT4 Q8 FP16	42 tok/s 29 tok/s 17 tok/s	~655 ms ~665 ms ~690 ms
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	357 tok/s 268 tok/s 172 tok/s	~60 ms ~61 ms ~63 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	357 tok/s 268 tok/s 172 tok/s	~60 ms ~61 ms ~63 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	345 tok/s 293 tok/s 214 tok/s	~38 ms ~38 ms ~40 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	345 tok/s 293 tok/s 214 tok/s	~38 ms ~38 ms ~40 ms
-	Falcon-1B	1B	INT4 Q8 FP16	258 tok/s 173 tok/s 101 tok/s	~122 ms ~124 ms ~128 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	237 tok/s 164 tok/s 98 tok/s	~122 ms ~124 ms ~129 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	182 tok/s 120 tok/s 69 tok/s	~178 ms ~181 ms ~187 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	182 tok/s 120 tok/s 69 tok/s	~178 ms ~181 ms ~187 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	166 tok/s 109 tok/s 61 tok/s	~201 ms ~204 ms ~211 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	160 tok/s 104 tok/s 58 tok/s	~212 ms ~215 ms ~223 ms
-	Falcon-3B	3B	INT4 Q8 FP16	118 tok/s 70 tok/s 37 tok/s	~337 ms ~342 ms ~355 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	115 tok/s 92 tok/s 62 tok/s	~159 ms ~161 ms ~167 ms
-	Phi-1	1.3B	INT4 Q8 FP16	115 tok/s 92 tok/s 62 tok/s	~159 ms ~161 ms ~167 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	115 tok/s 92 tok/s 62 tok/s	~159 ms ~161 ms ~167 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	111 tok/s 68 tok/s 37 tok/s	~337 ms ~343 ms ~355 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	110 tok/s 68 tok/s 37 tok/s	~337 ms ~343 ms ~355 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	110 tok/s 68 tok/s 37 tok/s	~337 ms ~343 ms ~355 ms
-	ERNIE-4.5-21B-A3B	21B (3B active)	INT4 Q8	92 tok/s 55 tok/s	~412 ms ~490 ms
-	ERNIE-4.5-21B-A3B-Base	21B (3B active)	INT4 Q8	92 tok/s 55 tok/s	~412 ms ~490 ms
-	ERNIE-4.5-VL-28B-A3B	28B (3B active)	INT4	88 tok/s	~433 ms
-	ERNIE-4.5-VL-28B-A3B-Base	28B (3B active)	INT4	88 tok/s	~433 ms
-	Hy-MT2 30B A3B	30B (3B active)	INT4	87 tok/s	~440 ms
-	Spark X2.5 4B	4B	INT4 Q8 FP16	87 tok/s 53 tok/s 28 tok/s	~441 ms ~448 ms ~465 ms
-	Phi-2	2.7B	INT4 Q8 FP16	71 tok/s 53 tok/s 33 tok/s	~312 ms ~317 ms ~328 ms
-	MaLLaM-3B	3B	INT4 Q8 FP16	67 tok/s 50 tok/s 31 tok/s	~343 ms ~348 ms ~361 ms
-	SmolLM3 3B	3B	INT4 Q8 FP16	67 tok/s 50 tok/s 31 tok/s	~343 ms ~348 ms ~361 ms
-	Falcon-7B	7B	INT4 Q8 FP16	57 tok/s 32 tok/s 17 tok/s	~749 ms ~763 ms ~792 ms
-	Falcon3-7B	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
-	Hy-MT2 7B	7B	INT4 Q8 FP16	54 tok/s 31 tok/s 16 tok/s	~750 ms ~763 ms ~793 ms
-	Mistral-7B-v0.1	7.3B	INT4 Q8 FP16	52 tok/s 30 tok/s 16 tok/s	~781 ms ~794 ms ~825 ms
-	LensVLM 9B	9B	INT4 Q8 FP16	43 tok/s 25 tok/s 12 tok/s	~958 ms ~975 ms ~1.1s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4 Q8 FP16	43 tok/s 25 tok/s 12 tok/s	~958 ms ~975 ms ~1.1s
-	ChatGLM3-6B-32K	6B	INT4 Q8 FP16	42 tok/s 29 tok/s 17 tok/s	~655 ms ~665 ms ~690 ms
-	Yi-6B	6B	INT4 Q8 FP16	42 tok/s 29 tok/s 17 tok/s	~655 ms ~665 ms ~690 ms
-	Falcon3-10B	10B	INT4 Q8 FP16	39 tok/s 22 tok/s 11 tok/s	~1.1s ~1.1s ~1.2s
-	Kimi-VL-A3B-Instruct	16B (3B active)	INT4 Q8	39 tok/s 32 tok/s	~411 ms ~450 ms
-	Falcon2-11B	11B	INT4 Q8 FP16	37 tok/s 21 tok/s 10 tok/s	~1.2s ~1.2s ~1.4s
-	Hunyuan Lite	7B	INT4 Q8 FP16	37 tok/s 25 tok/s 15 tok/s	~759 ms ~771 ms ~800 ms
-	OLMo 3 7B Base	7B	INT4 Q8 FP16	37 tok/s 25 tok/s 15 tok/s	~759 ms ~771 ms ~800 ms
-	SEA-LION-7B	7.1B	INT4 Q8 FP16	37 tok/s 25 tok/s 14 tok/s	~769 ms ~782 ms ~811 ms
-	SEA-LION-7B-Instruct	7.1B	INT4 Q8 FP16	37 tok/s 25 tok/s 14 tok/s	~769 ms ~782 ms ~811 ms
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4 Q8 FP16	33 tok/s 22 tok/s 12 tok/s	~862 ms ~877 ms ~986 ms
-	Typhoon-2-8B	8B	INT4 Q8 FP16	33 tok/s 22 tok/s 12 tok/s	~862 ms ~877 ms ~986 ms
-	GLM-4-9B	9B	INT4 Q8 FP16	30 tok/s 20 tok/s 11 tok/s	~968 ms ~984 ms ~1.1s
-	GLM-4-9B-Chat	9B	INT4 Q8 FP16	30 tok/s 20 tok/s 11 tok/s	~968 ms ~984 ms ~1.1s
-	GLM-4-9B-Chat-1M	9B	INT4 Q8 FP16	30 tok/s 20 tok/s 11 tok/s	~968 ms ~984 ms ~1.1s
-	Yi-9B	9B	INT4 Q8 FP16	30 tok/s 20 tok/s 11 tok/s	~968 ms ~984 ms ~1.1s
-	Sahabat-AI-Gemma2-9B	9.2B	INT4 Q8 FP16	29 tok/s 20 tok/s 11 tok/s	~989 ms ~1.0s ~1.1s
-	Sahabat-AI-Gemma2-9B-Instruct	9.2B	INT4 Q8 FP16	29 tok/s 20 tok/s 11 tok/s	~989 ms ~1.0s ~1.1s
-	GreenMind-14B-R1	14B	INT4 Q8	20 tok/s 13 tok/s	~1.5s ~1.5s
-	Falcon-40B	40B	INT4	10 tok/s	~4.5s

NVIDIA L4 (24GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA L4 (24GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions