Skip to content

What can the RTX 3080 10 GB run?

What fits on the RTX 3080 10 GB
Text models: 244Visual models: 12Strongest first
Memory
10 GB
Bandwidth
760 GB/s
Maker
NVIDIA

Open the calculator on this machine

Text models that run on it

Text models that run on the RTX 3080 10 GB, best first
ModelBest fitContextSpeedStrongest at
Qwen3.5 9B6-bit32K70 tok/sChat, Coding, Agents, Docs
Gemma 4 12B5-bit24K57 tok/sChat, Coding, Docs
Qwen3.5 4B8-bit120K96 tok/sAgents, Docs
Gemma 4 E4B8-bit32K86 tok/sChat, Coding, Docs
Ministral 3 8B5-bit20K82 tok/sDocs
Gemma 4 E2B8-bit128K152 tok/sChat
Nemotron 3 Nano 4B8-bit256K117 tok/sDocs
Llama 3.1 8B6-bit20K80 tok/sChat, Coding, Agents, Docs
Qwen3 0.6B8-bit40K427 tok/sNot rated
Qwen3 VL 8B Instruct5-bit20K84 tok/sText and images
Qwen3 8B6-bit16K78 tok/sNot rated
Qwen2.5 7B Instruct8-bit20K67 tok/sNot rated
Qwen2.5 VL 7B Instruct6-bit40K78 tok/sText and images
Unlimited OCR8-bit32K333 tok/sReading documents
Qwen2.5 1.5B Instruct8-bit32K255 tok/sNot rated
Qwen3.5 2B8-bit256K189 tok/sText and images
Qwen3 VL 4B Instruct8-bit28K101 tok/sText and images
Qwen3 1.7B8-bit40K204 tok/sNot rated
Qwen3 4B8-bit32K110 tok/sNot rated
Qwen3 4B Instruct 25078-bit32K110 tok/sNot rated
Qwen2.5 3B Instruct8-bit32K138 tok/sNot rated
DeepSeek OCR8-bit8K333 tok/sReading documents
Qwen2.5 VL 3B Instruct8-bit125K117 tok/sText and images
Qwen2.5 Coder 7B Instruct8-bit20K67 tok/sCoding
Mistral 7B Instruct V0.26-bit24K84 tok/sNot rated
Chandra OCR 28-bit96K86 tok/sReading documents
Llama 3.2 1B Instruct8-bit128K339 tok/sNot rated
Qwen3 VL 2B Instruct8-bit60K196 tok/sText and images
Qwen2 VL 2B Instruct8-bit32K190 tok/sText and images
MiniCPM5 2B8-bit128K174 tok/sNot rated
Gemma 3 4B IT8-bit128K121 tok/sText and images
Llama 3.2 3B Instruct8-bit48K138 tok/sNot rated
LightOnOCR 2 1B8-bit16K351 tok/sText and images
Qwen3 1.7B Base8-bit32K234 tok/sNot rated
HunyuanOCR8-bit80K338 tok/sText and images
Qwen3 0.6B Base8-bit32K493 tok/sNot rated
Rax 4.58-bit256K189 tok/sText and images
Qwen2 VL 7B Instruct6-bit32K78 tok/sText and images
Ornith 1.5 9B6-bit32K70 tok/sNot rated
Unlimited OCR 28-bit88K332 tok/sReading documents
Granite 4.1 3B8-bit64K125 tok/sNot rated
DeepSeek R1 Distill Qwen 1.5B8-bit128K255 tok/sNot rated
Mistral 7B V0.18-bit32K66 tok/sNot rated
Qwen2.5 Coder 3B Instruct8-bit32K138 tok/sCoding
Meta Llama 3 8B Instruct Abliterated V36-bit8K80 tok/sNot rated
EXAONE 3.5 7.8B Instruct6-bit20K81 tok/sNot rated
Saiga Llama3 8B6-bit8K80 tok/sNot rated
MiniCPM V 4.7 1B8-bit256K298 tok/sText and images
Qwen2.5 Coder 1.5B Instruct8-bit32K255 tok/sCoding
DeepSeek R1 0528 Qwen3 8B6-bit16K78 tok/sNot rated
Gemma 3 12B IT4-bit16K71 tok/sText and images
Phi 3.5 Mini Instruct5-bit16K168 tok/sNot rated
Qwen3 4B Thinking 25078-bit32K110 tok/sNot rated
Phi 4 Mini Instruct8-bit36K116 tok/sNot rated
Dots.ocr8-bit128K155 tok/sReading documents
DeepSeek R1 Distill Qwen 7B8-bit20K67 tok/sNot rated
Phi 3 Mini 4K Instruct8-bit4K119 tok/sNot rated
DeepSeek OCR 28-bit8K322 tok/sReading documents
SmolLM3 3B8-bit64K138 tok/sNot rated
Qwen2 7B Instruct8-bit20K67 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 244 are shown.

Image and video models

Image and video models that run on the RTX 3080 10 GB
ModelBest fitPer outputStrongest at
Stable Diffusion XL 1.016-bit11 sImages
Z-Image Turbo8-bit16 sImages
FLUX.1 [dev]5-bit36 sImages
Qwen-Image 2.18-bit4.2 minImages
FLUX.3 Klein8-bit23 sImages
FLUX.2 [klein] 9B8-bit7.9 sImages
Krea 2 Turbo5-bit35 sImages
Stable Diffusion 3.5 Large8-bit2.2 minImages
MOSS-Image8-bit2.7 minImages
Wan 2.2 TI2V-5B8-bit23 minVideo
MOSS Image Base8-bit2.7 minImages
HunyuanVideo 1.55-bit2.0 hVideo

Only just too big

Questions

What is the best LLM for the RTX 3080 10 GB?

Of the open models we track, the strongest that run well on the RTX 3080 10 GB are Qwen3.5 9B, Gemma 4 12B, Qwen3.5 4B. 244 text models fit it in all.

Calculated from published specs, not benchmarked. All machines