What are the best models to run on NVIDIA RTX 4060 (8GB)?

Based on model rankings and 8 GB usable VRAM: Top Overall: K2 Horizon 7B (INT4) ~49 tok/s. Top Coding: Qwen3.5-9B (INT4) ~39 tok/s. Optimal Daily Driver: Granite 4.2 8B (INT4) ~44 tok/s. Maximum Model Size: Gemma 3 12B (INT4) ~27 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 4060 (8GB) (October 2026)

Rank	Model	Parameters	Precision	TPS (Approx.)	TTFT
#116	K2 Horizon 7B	7B	INT4	49 tok/s	~1.3s
#132	Qwen3.5-4B	4B	INT4 Q8	77 tok/s 48 tok/s	~768 ms ~776 ms
#134	Qwen3.5-9B	9B	INT4	39 tok/s	~1.7s
#139	NVIDIA Nemotron 3 Nano 30B-A3B	3.5B (30B active)	INT4 Q8	12 tok/s 7 tok/s	~5.3s ~5.8s
#147	Granite 4.2 8B	8B	INT4	44 tok/s	~1.5s
#153	Gemma 4 E4B	8B	INT4	44 tok/s	~1.5s
#177	Gemma 3 12B	12B	INT4	27 tok/s	~2.6s
#183	Ministral 3 8B	8B	INT4	28 tok/s	~1.8s
#184	OLMo 3 7B Think	7B	INT4	32 tok/s	~1.4s
#185	Qwen3.5-2B	2B	INT4 Q8 FP16	122 tok/s 83 tok/s 48 tok/s	~410 ms ~414 ms ~423 ms
#186	MiniCPM5-2B	2B	INT4 Q8 FP16	122 tok/s 83 tok/s 48 tok/s	~410 ms ~414 ms ~423 ms
#189	Gemma 4 E2B	5.1B	INT4 Q8	63 tok/s 37 tok/s	~965 ms ~1.1s
#192	Ministral 3 3B	3B	INT4 Q8	60 tok/s 45 tok/s	~594 ms ~600 ms
#193	DeepSeek-R1 8B	8B	INT4	45 tok/s	~1.5s
#194	Gemma 2 9B	9B	INT4	39 tok/s	~1.7s
#197	Gemma 3 4B	4B	INT4 Q8	77 tok/s 48 tok/s	~768 ms ~776 ms
#199	Qwen2.5-7B	7B	INT4	49 tok/s	~1.3s
#200	Qwen3-4B	4B	INT4 Q8	77 tok/s 48 tok/s	~768 ms ~776 ms
#201	DeepSeek-R1 7B	7B	INT4 Q8	50 tok/s 27 tok/s	~1.3s ~1.5s
#202	Qwen3-8B	8B	INT4	44 tok/s	~1.5s
#203	Llama 3.1 8B	8B	INT4	44 tok/s	~1.5s
#207	Llama 3 8B	8B	INT4	44 tok/s	~1.5s
#208	Gemma 3 270M	0.27B	INT4 Q8 FP16	251 tok/s 224 tok/s 178 tok/s	~61 ms ~61 ms ~62 ms
#209	Qwen3-1.7B	1.7B	INT4 Q8 FP16	135 tok/s 94 tok/s 55 tok/s	~352 ms ~355 ms ~362 ms
#210	Qwen3-0.6B	0.6B	INT4 Q8 FP16	231 tok/s 182 tok/s 124 tok/s	~127 ms ~128 ms ~131 ms
#211	Phi-4-Mini	3.8B	INT4 Q8	79 tok/s 50 tok/s	~732 ms ~739 ms
#212	Ministral-8B-2410	8B	INT4	44 tok/s	~1.5s
#214	DeepSeek-R1 1.5B	1.5B	INT4 Q8 FP16	146 tok/s 102 tok/s 62 tok/s	~312 ms ~315 ms ~321 ms
#215	Llama 3.2 3B	3B	INT4 Q8 FP16	94 tok/s 61 tok/s 32 tok/s	~588 ms ~594 ms ~658 ms
#216	Phi-3-mini	3.8B	INT4 Q8	79 tok/s 50 tok/s	~732 ms ~739 ms
#217	Qwen3.5-0.8B	0.8B	INT4 Q8 FP16	203 tok/s 155 tok/s 101 tok/s	~170 ms ~172 ms ~175 ms
#218	Gemma 3n E2B IT	6B (2B active)	INT4	57 tok/s	~483 ms
#219	Gemma 2 2B	2B	INT4 Q8 FP16	122 tok/s 83 tok/s 48 tok/s	~410 ms ~414 ms ~423 ms
#221	Mistral-7B-Instruct-v0.1	7.3B	INT4	47 tok/s	~1.4s
#222	Llama 3.2 1B	1B	INT4 Q8 FP16	181 tok/s 134 tok/s 85 tok/s	~215 ms ~217 ms ~221 ms
#224	OLMo 3 7B Instruct	7B	INT4	32 tok/s	~1.4s
#225	Qwen2-7B	7B	INT4	49 tok/s	~1.3s
#226	Phi-3-small	7B	INT4	49 tok/s	~1.3s
#227	Mistral-7B-Instruct-v0.2	7.3B	INT4	47 tok/s	~1.4s
#229	Gemma 3 1B	1B	INT4 Q8 FP16	181 tok/s 134 tok/s 85 tok/s	~215 ms ~217 ms ~221 ms
#230	Gemma 1 7B	7B	INT4	32 tok/s	~1.4s
#231	Gemma 1 2B	2B	INT4 Q8 FP16	132 tok/s 87 tok/s 50 tok/s	~410 ms ~413 ms ~422 ms
#232	ChatGLM3-6B	6B	INT4	36 tok/s	~1.2s
#233	ChatGLM2-6B	6B	INT4	36 tok/s	~1.2s
#234	ChatGLM-6B	6B	INT4	36 tok/s	~1.2s
Unranked Models (Sorted by Speed)
-	Qwen2-0.5B	0.5B	INT4 Q8 FP16	247 tok/s 199 tok/s 139 tok/s	~106 ms ~107 ms ~110 ms
-	Qwen2.5-0.5B	0.5B	INT4 Q8 FP16	247 tok/s 199 tok/s 139 tok/s	~106 ms ~107 ms ~110 ms
-	ERNIE-4.5-0.3B	0.3B	INT4 Q8 FP16	241 tok/s 213 tok/s 167 tok/s	~67 ms ~67 ms ~69 ms
-	ERNIE-4.5-0.3B-Base	0.3B	INT4 Q8 FP16	241 tok/s 213 tok/s 167 tok/s	~67 ms ~67 ms ~69 ms
-	Falcon-1B	1B	INT4 Q8 FP16	194 tok/s 140 tok/s 87 tok/s	~214 ms ~216 ms ~221 ms
-	Falcon3-1B	1B	INT4 Q8 FP16	181 tok/s 134 tok/s 85 tok/s	~215 ms ~217 ms ~221 ms
-	Qwen2-1.5B	1.5B	INT4 Q8 FP16	146 tok/s 102 tok/s 62 tok/s	~312 ms ~315 ms ~321 ms
-	Qwen2.5-1.5B	1.5B	INT4 Q8 FP16	146 tok/s 102 tok/s 62 tok/s	~312 ms ~315 ms ~321 ms
-	Spark X2.5 1.7B	1.7B	INT4 Q8 FP16	135 tok/s 94 tok/s 55 tok/s	~352 ms ~355 ms ~362 ms
-	Hy-MT2 1.8B	1.8B	INT4 Q8 FP16	131 tok/s 90 tok/s 53 tok/s	~370 ms ~373 ms ~381 ms
-	Falcon-3B	3B	INT4 Q8 FP16	100 tok/s 63 tok/s 33 tok/s	~588 ms ~594 ms ~658 ms
-	CroissantLLM Base	1.3B	INT4 Q8 FP16	98 tok/s 81 tok/s 56 tok/s	~276 ms ~278 ms ~284 ms
-	Phi-1	1.3B	INT4 Q8 FP16	98 tok/s 81 tok/s 56 tok/s	~276 ms ~278 ms ~284 ms
-	Phi-1.5	1.3B	INT4 Q8 FP16	98 tok/s 81 tok/s 56 tok/s	~276 ms ~278 ms ~284 ms
-	DeepSeek-R1 3B	3B	INT4 Q8 FP16	96 tok/s 61 tok/s 32 tok/s	~588 ms ~594 ms ~658 ms
-	Falcon3-3B	3B	INT4 Q8 FP16	94 tok/s 61 tok/s 32 tok/s	~588 ms ~594 ms ~658 ms
-	Qwen2.5-3B	3B	INT4 Q8 FP16	94 tok/s 61 tok/s 32 tok/s	~588 ms ~594 ms ~658 ms
-	FrogNano 4B	4B	INT4 Q8	77 tok/s 48 tok/s	~768 ms ~776 ms
-	Spark X2.5 4B	4B	INT4 Q8	77 tok/s 48 tok/s	~768 ms ~776 ms
-	Phi-2	2.7B	INT4 Q8 FP16	63 tok/s 49 tok/s 28 tok/s	~541 ms ~546 ms ~652 ms
-	MaLLaM-3B	3B	INT4 Q8	60 tok/s 45 tok/s	~594 ms ~600 ms
-	SmolLM3 3B	3B	INT4 Q8	60 tok/s 45 tok/s	~594 ms ~600 ms
-	Falcon-7B	7B	INT4 Q8	51 tok/s 27 tok/s	~1.3s ~1.5s
-	Falcon3-7B	7B	INT4	49 tok/s	~1.3s
-	Hy-MT2 7B	7B	INT4	49 tok/s	~1.3s
-	Mistral-7B-v0.1	7.3B	INT4	47 tok/s	~1.4s
-	LensVLM 9B	9B	INT4	39 tok/s	~1.7s
-	MiMo V2.6 Distill Qwen 9B	9B	INT4	39 tok/s	~1.7s
-	ChatGLM3-6B-32K	6B	INT4	36 tok/s	~1.2s
-	Yi-6B	6B	INT4	36 tok/s	~1.2s
-	Falcon3-10B	10B	INT4	34 tok/s	~2.0s
-	Falcon2-11B	11B	INT4	33 tok/s	~2.2s
-	Hunyuan Lite	7B	INT4	32 tok/s	~1.4s
-	OLMo 3 7B Base	7B	INT4	32 tok/s	~1.4s
-	SEA-LION-7B	7.1B	INT4	32 tok/s	~1.4s
-	SEA-LION-7B-Instruct	7.1B	INT4	32 tok/s	~1.4s
-	Sahabat-AI-Llama3-8B-Instruct	8B	INT4	28 tok/s	~1.8s
-	Typhoon-2-8B	8B	INT4	28 tok/s	~1.8s

NVIDIA RTX 4060 (8GB)

VRAM Scaling by Context Length

Best Models to Run on NVIDIA RTX 4060 (8GB)

Inference Capacity

Fine-Tuning Capacity

Workload Recommendations

Frequently Asked Questions