AI models that fit on a 12 GB GPU

Open models that fit in 12 GB of VRAM: 125 as published, 267 at FP8 and 58 at INT4. Each model is listed on the smallest tier it fits at that precision, so anything smaller is on the 8 GB page or below.

GPUs with 12 GB

Cards whose datasheet VRAM puts them in this tier, up to the next tier at 16 GB. Prices are the lowest live data-center on-demand rate per GPU.

GPUVRAMFrom per GPU hour
Radeon RX 6700 XT12 GBNo live offer
RTX 3080 Ti12 GBNo live offer
RTX 407012 GBNo live offer
RTX 4070 Ti12 GBNo live offer
RTX 507012 GBNo live offer

Fits as published

Models whose published weights, plus the flat overhead, fit this much memory with no quantization.

Text models

The 40 most downloaded of 48.

ModelParametersPublished asVRAM needed
Qwen3-4B4.0BBF169.0 GB
Qwen3-4B-Instruct-25074.0BBF169.0 GB
Qwen3-4B-Base4.0BBF169.0 GB
Qwen3-8B-FP88.2BF8_E4M39.2 GB
Phi-3-mini-4k-instruct3.8BBF168.5 GB
Phi-4-mini-instruct3.8BBF168.6 GB
Qwen3-8B-FP88.2BF8_E4M39.2 GB
NVIDIA-Nemotron-3-Nano-4B-BF164.0BBF168.9 GB
Qwen3-4B-Thinking-25074.0BBF169.0 GB
Agents-A1-4B4.5BBF1610.1 GB
Phi-3.5-mini-instruct3.8BBF168.5 GB
gemma-2-2b2.6BF3211.7 GB
Phi-3-mini-128k-instruct3.8BBF168.5 GB
VLM2Vec-Full4.1BBF169.3 GB
Qwen3Guard-Gen-4B4.4BBF169.9 GB
typhoon2.5-qwen3-4b4.0BBF169.0 GB
Meta-Llama-3.1-8B-Instruct-FP88.0BF8_E4M39.0 GB
NVIDIA-Nemotron-Nano-9B-v2-FP88.9BF8_E4M39.9 GB
Phi-3-vision-128k-instruct4.1BBF169.3 GB
EXAONE-3.5-2.4B-Instruct2.4BF3210.8 GB
smol-tools-4b-32k5.2BBF1611.6 GB
Phi-4-mini-reasoning3.8BBF168.6 GB
gemma-3-4B-T1-it3.9BBF168.7 GB
Nanbeige4.2-3B4.2BBF169.3 GB
Nemotron-Labs-Diffusion-3B3.8BBF168.6 GB
Llama-3.1-8B-Instruct-FP88.0BF8_E4M39.0 GB
Meta-Llama-3.1-8B-FP88.0BF8_E4M39.0 GB
Meta-Llama-3.1-8B-Instruct-FP8-dynamic8.0BF8_E4M39.0 GB
Spark-X2.6-4B4.1BBF169.2 GB
Qwen1.5-4B4.0BBF168.8 GB
granite-4.1-8b-fp88.8BF8_E4M39.8 GB
Nanbeige4.1-3B3.9BBF168.8 GB
Qwen1.5-4B-Chat4.0BBF168.8 GB
Qwen3-8B-FP8-dynamic8.2BF8_E4M39.2 GB
Ornith-1.5-9B-INT4-W4A16-AutoRound4.3BBF169.7 GB
NVIDIA-Nemotron-Labs-Dory1.0-4B-BF164.0BBF168.9 GB
Qwen3.8-4B-Distill4.7BBF1610.4 GB
granite-4.2-3b3.7BBF168.2 GB
AFM-4.5B4.6BBF1610.3 GB
DeepSeek-V4-Pro-Qwen3.5-4B-UNCENSORED-heretic4.5BBF1610.1 GB

Vision-language models

28 models.

ModelParametersPublished asVRAM needed
Qwen3.5-4B4.7BBF1610.4 GB
Qwen3-VL-4B-Instruct4.4BBF169.9 GB
Qwen2.5-VL-3B-Instruct3.8BBF168.4 GB
chandra-ocr-25.3BBF1611.8 GB
Qwen3-VL-8B-Instruct-FP88.8BF8_E4M39.8 GB
gemma-3-4b-it4.3BBF169.6 GB
medgemma-4b-it4.3BBF169.6 GB
Phi-3.5-vision-instruct4.1BBF169.3 GB
Qwen3.5-4B-Base4.7BBF1610.4 GB
typhoon-ocr-3b3.8BBF168.4 GB
Qwen3.5-9B-FP8-dynamic9.4BF8_E4M310.5 GB
Nanonets-OCR-s3.8BBF168.4 GB
vllm-translategemma-4b-it5.0BBF1611.1 GB
medgemma-1.5-4b-it4.3BBF169.6 GB
olmOCR-2-7B-1025-FP88.3BF8_E4M39.3 GB
InternVL2_5-4B3.7BBF168.3 GB
SmolVLM2-2.2B-Instruct2.2BF3210.0 GB
granite-vision-4.1-4b4.0BBF168.9 GB
Mage-VL4.7BBF1610.6 GB
LocateAnything-3B3.8BBF168.6 GB
Qwen3.5-4B-AWQ4.7BBF1610.4 GB
gemma-4-E2B-it-unsloth-bnb-4bit5.1BBF1611.5 GB
gemma-4-E4B-it-W4A16-AutoRound-GPTQ4.0BBF168.9 GB
Ovis1.6-Llama3.2-3B4.1BBF169.3 GB
WeVisOne-VL-4B4.8BBF1610.8 GB
Ovis2-4B4.6BBF1610.3 GB
Mage-VL-Cowork4.7BBF1610.6 GB
S5-nano4.7BBF1610.4 GB

Image generation models

17 models.

ModelParametersPublished asVRAM needed
stable-diffusion-xl-base-1.02.6BF3211.5 GB
sdxl-turbo2.6BF3211.5 GB
RealVisXL_V5.02.6BF3211.5 GB
playground-v2.5-1024px-aesthetic2.6BF3211.5 GB
stable-diffusion-xl-1.0-inpainting-0.12.6BF3211.5 GB
ideogram-4-fp89.3BF8_E4M310.4 GB
minecraft-skin-generator-sdxl2.6BF3211.5 GB
dreamshaper-xl-v2-turbo2.6BF3211.5 GB
NSFW-GEN-ANIME-v22.6BF3211.5 GB
LUSTIFY-v2.02.6BF3211.5 GB
MOSS-Image-SFT5.1BBF1611.4 GB
LUSTIFY-v2.02.6BF3211.5 GB
NSFW-Uncensored2.6BF3211.5 GB
pornmasterPro_noobV3VAE2.6BF3211.5 GB
MOSS-Image-Base5.1BBF1611.4 GB
dreamshaper-xl-lightning2.6BF3211.5 GB
pjit-21k2.2BF329.8 GB

Video generation models

4 models.

ModelParametersPublished asVRAM needed
LTX-Video1.9BF328.6 GB
FastWan2.2-TI2V-5B-FullAttn-Diffusers5.0BBF1611.2 GB
LongLive-2.0-5B-Diffusers5.0BBF1611.2 GB
Wan2.1-VACE-1.3B-diffusers2.2BF329.6 GB

Speech models

11 models.

ModelParametersPublished asVRAM needed
Voxtral-Mini-4B-Realtime-26024.4BBF169.9 GB
seamless-m4t-v2-large2.3BF3210.3 GB
s2-pro4.6BBF1610.2 GB
higgs-tts-3-4b4.7BBF1610.4 GB
Audio8-ASR-Infinite4.1BBF169.1 GB
YuE2-3B3.6BBF168.1 GB
MiniMax-Music32.4BF3210.9 GB
ARK-ASR-3B4.1BBF169.1 GB
Veena3.8BBF168.5 GB
parler-tts-large-v12.3BF3210.4 GB
acestep-5Hz-lm-4B4.2BBF169.4 GB

Embedding models

2 models.

ModelParametersPublished asVRAM needed
Qwen3-Embedding-4B4.0BBF169.0 GB
Qwen3-VL-Embedding-8B-FP88.8BF8_E4M39.8 GB

Other models

15 models.

ModelParametersPublished asVRAM needed
gemma-4-E2B-it5.1BBF1611.5 GB
Qwen3-Reranker-4B4.0BBF169.0 GB
FLUX.2-klein-4B3.9BBF168.7 GB
bge-reranker-v2-gemma2.5BF3211.2 GB
Qwen3.5-4B-3D-Coder4.5BF1610.1 GB
FLUX.2-klein-base-4B3.9BBF168.7 GB
NuExtract34.5BBF1610.1 GB
Ripple-65B-Preview-DFlash24.0BBF169.0 GB
Haru-4B5.3BBF1611.8 GB
EVIE-Preview-4.5B4.5BBF1610.1 GB
pplx-decider-flash-v1.34.5BBF1610.1 GB
FrogNano-4B-26094.7BBF1610.4 GB
kanana-2-v-decide-3b-26113.7BBF168.2 GB
THX-024.5BBF1610.1 GB
LightNav-04.8BBF1610.8 GB

Fits at FP8

Models that fit only after quantizing the weights to 8 bits (1 byte per parameter). Needs an FP8 checkpoint or an engine that quantizes on load, and a GPU with FP8 support.

Text models

The 40 most downloaded of 190.

ModelParametersPublished asVRAM needed
Qwen3-8B8.2BBF169.2 GB
Qwen2.5-7B-Instruct7.6BBF168.5 GB
Llama-3.1-8B-Instruct8.0BBF169.0 GB
Qwen2.5-Coder-7B-Instruct7.6BBF168.5 GB
Meta-Llama-3-8B-Instruct8.0BBF169.0 GB
Mistral-7B-Instruct-v0.27.2BBF168.1 GB
granite-4.1-8b8.8BBF169.8 GB
DeepSeek-R1-0528-Qwen3-8B8.2BBF169.2 GB
gemma-2-9b-it9.2BBF1610.3 GB
Meta-Llama-3-8B8.0BBF169.0 GB
Apertus-8B-Instruct-25098.1BBF169.0 GB
Qwen2.5-7B7.6BBF168.5 GB
Llama-3.1-8B8.0BBF169.0 GB
Qwen2.5-Coder-7B7.6BBF168.5 GB
falcon-7b7.2BBF168.1 GB
NVIDIA-Nemotron-Nano-9B-v28.9BBF169.9 GB
Qwen3-8B-Base8.2BBF169.2 GB
Qwen2-7B-Instruct7.6BBF168.5 GB
saiga_llama3_8b8.0BBF169.0 GB
Hermes-3-Llama-3.1-8B8.0BBF169.0 GB
DeepSeek-R1-Distill-Llama-8B8.0BBF169.0 GB
gemma4-e4b-claims-comparison7.9BBF168.9 GB
DeepSeek-R1-Distill-Qwen-7B7.6BBF168.5 GB
EXAONE-3.5-7.8B-Instruct7.8BF328.7 GB
Meta-Llama-3.1-8B-Instruct8.0BBF169.0 GB
Mistral-7B-v0.17.2BBF168.1 GB
Meta-Llama-3.1-8B-Instruct8.0BBF169.0 GB
Olmo-3-7B-Instruct7.3BBF168.2 GB
Fanar-1-9B-Instruct8.8BBF169.8 GB
LLaDA-8B-Instruct8.0BBF169.0 GB
Qwen2-7B7.6BBF168.5 GB
Ornith-1.5-9B9.7BBF1610.8 GB
Llama-3.1-Nemotron-Nano-8B-v18.0BBF169.0 GB
granite-guardian-3.3-8b8.2BBF169.1 GB
Llama-Guard-3-8B8.0BBF169.0 GB
Qwen2.5-7B-Instruct-1M7.6BBF168.5 GB
Mistral-7B-Instruct-v0.17.2BBF168.1 GB
Nemotron-Labs-Diffusion-8B8.5BBF169.5 GB
Llama-3-Taiwan-8B-Instruct8.0BBF169.0 GB
granite-3.0-8b-instruct8.2BBF169.1 GB

Vision-language models

The 40 most downloaded of 43.

ModelParametersPublished asVRAM needed
Qwen3.5-9B9.7BBF1610.8 GB
Qwen3-VL-8B-Instruct8.8BBF169.8 GB
Qwen2.5-VL-7B-Instruct8.3BBF169.3 GB
Qwen2-VL-7B-Instruct8.3BBF169.3 GB
Qwen3.5-9B-AWQ9.7BBF1610.8 GB
UI-TARS-1.5-7B8.3BF329.3 GB
llava-v1.6-mistral-7b-hf7.6BF168.5 GB
Cosmos-Reason2-8B8.8BBF169.8 GB
gemma-4-E4B-it-unsloth-bnb-4bit8.0BBF168.9 GB
Qwen3.5-9B-Base9.7BBF1610.8 GB
Ovis2.5-9B9.2BBF1610.3 GB
qwen3vl-resume-parser8.8BBF169.8 GB
MiniCPM-V-4_58.7BBF169.7 GB
ChatRex-7B7.3BBF168.2 GB
Llama-3.1-Nemotron-Nano-VL-8B-V18.7BBF169.7 GB
Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED9.4BBF1610.5 GB
GLM-4.1V-9B-Thinking10.3BBF1611.5 GB
olmOCR-2-7B-10258.3BBF169.3 GB
moondream3-preview9.3BBF1610.4 GB
lift9.7BBF1610.8 GB
gemma-3n-E4B-it8.4BBF169.4 GB
Molmo2-8B8.7BF329.7 GB
idefics2-8b8.4BF329.4 GB
Llama-3.2-11B-Vision-Instruct10.7BBF1611.9 GB
Idefics3-8B-Llama38.5BBF169.5 GB
Cosmos-Reason1-7B8.3BBF169.3 GB
InternVL3-8B7.9BBF168.9 GB
GLM-4.6V-Flash10.3BBF1611.5 GB
Qwen3-VL-8B-Thinking8.8BBF169.8 GB
XCurOS1.2-8B-VLBF16-Instruct8.8BBF169.8 GB
llava-v1.6-mistral-7b7.6BBF168.5 GB
gemma-3n-E4B-it-MLX-bf167.8BBF168.8 GB
Molmo-7B-D-09248.0BF329.0 GB
gemma-3n-E4B-it7.8BBF168.8 GB
deepseek-vl-7b-chat7.3BF168.2 GB
MiMo-V2.6-Distill-Qwen-9B9.4BBF1610.5 GB
Ovis1.6-Gemma2-9B10.2BBF1611.4 GB
llava-med-v1.5-mistral-7b7.6BBF168.5 GB
ZDTaichu5.0-9B9.8BBF1610.9 GB
clef-flash9.4BBF1610.5 GB

Image generation models

6 models.

ModelParametersPublished asVRAM needed
stable-diffusion-3.5-large8.1BBF169.1 GB
Chroma1-HD8.9BBF169.9 GB
Chroma1-Base8.9BBF169.9 GB
FIBO8.3BBF169.3 GB
flux2-klein-9b-uncensored-text-encoder8.2BF169.2 GB
Fibo-1.58.3BBF169.3 GB

Video generation models

1 model.

ModelParametersPublished asVRAM needed
mochi-1-preview10.0BF3211.2 GB

Speech models

4 models.

ModelParametersPublished asVRAM needed
VibeVoice-ASR8.7BBF169.7 GB
granite-speech-3.2-8b8.5BBF169.5 GB
granite-speech-3.3-8b8.6BBF169.7 GB
higgs-audio-v3-8b-stt-v28.9BBF1610.0 GB

Embedding models

5 models.

ModelParametersPublished asVRAM needed
Qwen3-Embedding-8B7.6BBF168.5 GB
Qwen3-VL-Embedding-8B8.1BBF169.1 GB
gte-Qwen2-7B-instruct7.6BF328.5 GB
WeMM-Embedding-9B9.4BBF1610.5 GB
pplx-embed-v2-context-9b-preview8.4BF329.4 GB

Other models

18 models.

ModelParametersPublished asVRAM needed
gemma-4-E4B-it8.0BBF168.9 GB
gemma-4-E4B-it-qat-w4a16-ct8.7BBF169.8 GB
gemma-4-E4B8.0BBF168.9 GB
gemma-4-E4B-it-uncensored-heretic8.0BBF168.9 GB
NuMarkdown-8B-Thinking8.3BBF169.3 GB
JEV-9B9.0BBF1610.0 GB
FLUX.2-klein-9B9.1BBF1610.1 GB
gemma-4-E4B-it-AWQ-INT48.2BF169.1 GB
personaplex-7b-v18.4BBF169.4 GB
bge-reranker-v2.5-gemma2-lightweight9.2BF3210.3 GB
Qwen3-Reranker-8B8.2BBF169.2 GB
openvla-7b7.5BBF168.4 GB
VibeVoice-ASR-HF8.3BBF169.3 GB
LLaVA-Video-7B-Qwen28.0BBF169.0 GB
rank1-7b7.6BBF168.5 GB
elephant-19.4BBF1610.5 GB
Fibo-2.0-edit-turbo8.1BBF169.1 GB
BluePixel-Turbo9.1BF3210.1 GB

Fits at INT4

Models that fit only after quantizing to 4 bits (0.5 byte per parameter). Needs a quantized checkpoint actually published for the model; check its Hugging Face page before relying on a row.

Text models

35 models.

ModelParametersPublished asVRAM needed
Qwen2.5-14B-Instruct14.8BBF168.3 GB
Gemma-4-31B-IT-NVFP420.9BBF1611.7 GB
Qwen2.5-Coder-14B-Instruct14.8BBF168.3 GB
Qwen3-14B14.8BBF168.3 GB
gpt-neox-20b20.7BF1611.6 GB
DeepSeek-Coder-V2-Lite-Instruct15.7BBF168.8 GB
Qwen3-14B-Base14.8BBF168.3 GB
phi-414.7BBF168.2 GB
DeepSeek-R1-Distill-Qwen-14B14.8BBF168.3 GB
DeepSeek-V2-Lite-Chat15.7BBF168.8 GB
DeepSeek-V2-Lite15.7BBF168.8 GB
LLaDA2.0-mini16.3BBF169.1 GB
Qwen2.5-Coder-14B14.8BBF168.3 GB
DeepSeek-Coder-V2-Lite-Instruct-FP815.7BF8_E4M38.8 GB
LLaDA2.1-mini16.3BBF169.1 GB
phi-4-FP8-dynamic14.7BF8_E4M38.2 GB
gpt-oss-20b-BF1620.9BBF1611.7 GB
Qwen3-14B-FP814.8BF8_E4M38.3 GB
Qwen2.5-14B14.8BBF168.3 GB
Qwen2.5-14B-Instruct-1M14.8BBF168.3 GB
Moonlight-16B-A3B-Instruct16.0BBF168.9 GB
JiRackUltra_14b14.8BBF168.3 GB
deepseek-moe-16b-chat16.4BBF169.2 GB
HyperCLOVAX-SEED-Think-14B14.7BF328.2 GB
Phi-4-reasoning14.7BBF168.2 GB
internlm2-chat-20b19.9BBF1611.1 GB
Moonlight-16B-A3B16.0BBF168.9 GB
internlm2-chat-20b-ExPO19.9BBF1611.1 GB
deepseek-moe-16b-base16.4BBF169.2 GB
Huihui-gpt-oss-20b-BF16-abliterated20.9BBF1611.7 GB
Param2-17B-A2.4B-Thinking17.2BBF169.6 GB
Ling-mini-2.016.3BBF169.1 GB
Phi-4-reasoning-plus14.7BBF168.2 GB
DeepSeek-R1-Distill-Qwen-14B-abliterated-v214.8BBF168.3 GB
ContextPilot-14B14.8BBF168.3 GB

Vision-language models

2 models.

ModelParametersPublished asVRAM needed
Qwen3.6-27B-NVFP421.2BF8_E4M311.9 GB
Kimi-VL-A3B-Instruct16.4BBF169.2 GB

Image generation models

4 models.

ModelParametersPublished asVRAM needed
Qwen-Image20.4BBF1611.4 GB
HiDream-I1-Fast17.1BBF169.6 GB
Qwen-Image-251220.4BBF1611.4 GB
HiDream-I1-Full17.1BF169.6 GB

Video generation models

12 models.

ModelParametersPublished asVRAM needed
LTX-218.9BBF1610.5 GB
Wan2.1-I2V-14B-720P-Diffusers16.4BF329.2 GB
Wan2.1-I2V-14B-480P-Diffusers16.4BF329.2 GB
LTX-2.5-Diffusers19.0BBF1610.6 GB
Wan2.1-I2V-14B-480P16.4BF329.2 GB
Wan2.1-I2V-14B-720P16.4BF329.2 GB
Wan2.2-S2V-14B16.3BBF169.1 GB
Wan2.2-S2V-14B-Diffusers16.3BBF169.1 GB
lingbot-world-fast18.5BF3210.4 GB
Cosmos-H-Surgical15.2BBF168.5 GB
Wan2.1-VACE-14B17.3BF329.7 GB
LTX-2.3-Diffusers19.0BBF1610.6 GB

Other models

5 models.

ModelParametersPublished asVRAM needed
Qwen3.8-27B-NVFP419.9BF8_E4M311.1 GB
Qwen-Image-Edit-251120.4BBF1611.4 GB
SenseNova-U1-8B-MoT17.6BBF169.8 GB
A.X-K2-Raon-Speech-21B-A3B21.2BBF1611.8 GB
SenseNova-U1.5-8B-MoT17.5BBF169.8 GB

How these numbers are computed

Required VRAM is the weight size at each precision times a flat 1.2 overhead, the same figure every model page shows. It does not include a long context: the KV cache grows with every token, so a model near the top of a tier can need the next one at long context. Read how much VRAM you need for LLMs for the method, the VRAM and quantization glossary entries for the terms, and the VRAM calculator to size a model that is not listed.

Looking for a model by job rather than by memory? Start with coding, reasoning, chat and assistants, vision-language or see the full models directory.

Other VRAM tiers

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.