§01·spec · /gpus

RTX 4090

Name: RTX 4090 local AI benchmarks
Creator: Smeltcore
License: https://creativecommons.org/licenses/by-sa/4.0/

nvidia40 series24GB VRAM

36 open-weights AI models run on the RTX 4090 — see which fit, how fast they go, and the VRAM each needs.

§02·models that run on this GPU

36 total

LLM · 15

Model	Best speed	Min VRAM	Works	Evidence
Llama 3.3 70B	17.79tokens/s	24GB	✓	1benchrecipe	check ↗
DeepSeek R1 Distill 14B		10GB	✓	recipe	check ↗
Devstral Small 2 (24B)		24GB	✓	recipe	check ↗
Gemma 4 12B		24GB	✓	recipe	check ↗
gpt-oss 20B		16GB	✓	recipe	check ↗
Laguna XS 2.1		24GB	✓	recipe	check ↗
Llama 3.1 8B		10GB	✓	recipe	check ↗
Mistral Nemo 12B		24GB	✓	recipe	check ↗
Mistral Small 3.2 24B		24GB	✓	recipe	check ↗
North Mini Code 1.0		24GB	✓	recipe	check ↗
Ornith 1.0 35B		24GB	✓	recipe	check ↗
Phi-4		24GB	✓	recipe	check ↗
Qwen3 14B		10GB	✓	recipe	check ↗
Qwen3 32B		22GB	✓	recipe	check ↗
Qwen3-8B		6GB	✓	recipe	check ↗

Multimodal · 2

Model	Best speed	Min VRAM	Works	Evidence
Gemma 4 E4B-IT		20GB	✓	recipe	check ↗
Gemma 4 26B MoE		18GB	✓	recipe	check ↗

Image · 9

Model	Best speed	Min VRAM	Works	Evidence
Z-Image Turbo	2.3s	13GB	✓	1benchrecipe	check ↗
Flux.1 Dev	1.85it/s	24.564GB	✓	1benchrecipe	check ↗
Chroma V48		24GB	✓	recipe	check ↗
Flux.2-Klein-4B		20GB	✓	1benchrecipe	check ↗
HiDream-O1-Image		20GB	✓	recipe	check ↗
Juggernaut Z		16GB	✓	recipe	check ↗
Krea 2		16GB	✓	recipe	check ↗
LongCat Image		18GB	✓	recipe	check ↗
Qwen-Image		21GB	✓	recipe	check ↗

Video · 9

Model	Best speed	Min VRAM	Works	Evidence
Wan 2.2 14B	4.33min	24GB	✓	1benchrecipe	check ↗
Wan 2.1	4min	8.19GB	✓	1benchrecipe	check ↗
CogVideoX 1.5		14GB	✓	recipe	check ↗
HunyuanVideo 1.5		14GB	✓	recipe	check ↗
LightX2V		24GB	✓	recipe	check ↗
LTX-2		24GB	✓	recipe	check ↗
LTX-2.3		20GB	✓	recipe	check ↗
Mochi 1		20GB	✓	recipe	check ↗
WAN 2.2		24GB	✓	recipe	check ↗

3D · 1

Model	Best speed	Min VRAM	Works	Evidence
Waypoint 1.5		12GB	✓	recipe	check ↗

§03·tested recipes

showing 6 of 38