RunsOnMy ModelsGPUs & MacsBy VRAMBest GPUCan I run…?OllamaStart hereAPI pricesVRAM calculatorNew

What AI models run on the RTX 3080 12GB?

The RTX 3080 12GB has 12 GB of VRAM (912 GB/s). About 11.4 GB is usable for a model. 173 of 298 open models we track fit at 8K context.

173models fit (8K ctx)
154still fit at 32K ctx
31Blargest that fits
912 GB/smemory bandwidth

This card has 12 GB of VRAM. See every model that fits in 12 GB VRAM. Other 12 GB hardware: RTX 3060 12GB, RTX 4070, RTX 4070 Super, RTX 4070 Ti, RTX 5070, RX 7700 XT, Arc B580.

Using a specific tool? Ollama models that run on the RTX 3080 12GB · LM Studio · llama.cpp offload

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)Details
GLM 4.7 Flash31B (4.6B act.)Q2_K11.4 GBno560details
Muse Glimmer 30B30BQ2_K11.4 GBno82details
Raon 1 Mini Preview29B (4.3B act.)Q2_K11.2 GBno563details
Arete 27B27BQ2_K11.2 GBno86details
MiniMax H3 Image PE27BQ2_K11.3 GBno85details
Trinity Mini26B (2.0B act.)Q2_K9.8 GBQ2_K1287details
GLiDE 27B26BQ2_K10.5 GBno93details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q2_K10.4 GBQ2_K608details
LFM2.5 24B A2B24B (2.0B act.)Q2_K9.2 GBQ2_K1216details
ERNIE 4.5 21B A3B PT22BQ2_K8.6 GBQ2_K113details
ERNIE 4.5 21B A3B Thinking22BQ2_K8.6 GBQ2_K113details
Kimi VL A3B Thinking 250616B (3.0B act.)Q3_K_M8.4 GBQ3_K_M622details
Kimi VL A3B Thinking16B (3.0B act.)Q3_K_M8.4 GBQ3_K_M622details
Kimi VL A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M602details
DeepSeek MoE 16B Base16B (2.8B act.)Q3_K_M10.2 GBno623details
LLaDA2.1 Mini16B (1.4B act.)Q3_K_M8.4 GBQ3_K_M1302details
Moonlight 16B A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M586details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q3_K_M8.0 GBQ3_K_M701details
DeepSeek V2 Lite16B (2.7B act.)Q4_K_M9.8 GBQ3_K_M561details
DeepSeek V2 Lite Chat16B (2.7B act.)Q3_K_M8.4 GBQ3_K_M662details
Qwen2.5 Coder 14B Instruct15BQ3_K_M9.0 GBno124details
DeepSeek R1 Distill Qwen 14B15BQ3_K_M9.0 GBno124details
Qwen3 14B15BQ3_K_M8.7 GBQ2_K126details
Phi 415BQ3_K_M9.0 GBno124details
Gemma 3 12B IT12BQ5_K_M9.6 GBQ4_K_M108details
Llama Guard 4 12B12BQ5_K_M10.2 GBQ2_K107details
KORMo Web 11B11BQ5_K_M9.3 GBQ2_K114details
Llama 3.2 11B Vision Instruct11BQ6_K10.0 GBQ2_K104details
Llama 3.2 11B Vision11BQ6_K10.0 GBQ2_K105details
GLM 4.6V Flash10BQ6_K8.7 GBQ6_K110details
GLM 4.1V 9B Thinking10BQ6_K8.7 GBQ6_K110details
Step3 VL 10B10BQ8_09.9 GBQ4_K_M105details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ6_K96details
Qwen3.5 9B9.7BQ8_010.3 GBQ6_K93details
Ornith 1.5 9B9.7BQ8_010.0 GBQ6_K96details
SenseNova Loop 9B9.4BQ8_010.0 GBQ6_K96details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ6_K96details
GLM 4 9B9.4BQ6_K8.7 GBQ6_K110details
Apertus v1.5 8B8.9BQ6_K8.4 GBQ4_K_M125details
InternLM3 8B Instruct8.8BQ8_09.7 GBQ6_K97details
Granite 4.2 8B8.8BQ6_K8.6 GBQ3_K_M126details
Granite 4.1 8B8.8BQ6_K8.6 GBQ3_K_M126details
Granite 4.1 8B Llmon Preview8.8BQ6_K8.6 GBQ3_K_M126details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ4_K_M105details
Molmo2 8B8.7BQ8_09.9 GBQ4_K_M105details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0857details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_0113details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_0113details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_0113details
olmOCR 7B 0225 Preview8.3BQ8_08.7 GBQ8_0113details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_0113details
Qwen3 8B8.2BQ8_09.9 GBQ4_K_M105details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ4_K_M105details
Granite 3.3 8B Instruct8.2BQ8_09.8 GBQ4_K_M105details
Granite 3.1 8B Instruct8.2BQ8_09.8 GBQ4_K_M105details
Granite 3.2 8B Instruct8.2BQ8_09.8 GBQ4_K_M105details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ5_K_M106details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M107details
Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M107details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M107details
Llama Guard 3 8B8.0BQ8_09.6 GBQ5_K_M107details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ5_K_M107details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ5_K_M107details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ5_K_M107details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M107details
Aya Expanse 8B8.0BQ8_09.8 GBQ4_K_M107details
Ling 3.0 Tiny7.9B (1.4B act.)Q8_08.7 GBQ8_0620details
MiMo 7B RL7.8BQ8_09.4 GBQ4_K_M113details
InternLM2.5 7B Chat7.7BQ8_09.2 GBQ5_K_M111details
InternLM2 Chat 7B7.7BQ8_09.2 GBQ5_K_M111details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_0113details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_0113details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_0113details
Olmo 3 7B Instruct7.3BQ6_K8.7 GBQ4_K_M152details
Olmo 3 1025 7B7.3BQ6_K8.7 GBQ4_K_M152details
Olmo 3 7B Think7.3BQ6_K8.7 GBQ4_K_M152details
Mistral 7B v0.17.2BQ8_08.6 GBQ6_K119details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ6_K119details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ6_K119details
Falcon 7B7.2BQ8_07.7 GBQ8_0119details
Falcon 7B Instruct7.2BQ8_07.4 GBQ8_0124details
Granite 4.0 H Tiny6.9B (1.5B act.)Q8_07.5 GBQ8_0584details
OLMoE 1B 7B 09246.9B (1.3B act.)Q8_08.4 GBQ6_K669details
DeepSeek Coder 7B Instruct v1.56.9BQ6_K9.6 GBno161details
DeepSeek Coder 6.7B Instruct6.7BQ6_K9.6 GBno165details
DeepSeek Coder 6.7B Base6.7BQ6_K9.6 GBno165details
Llama 2 7B HF6.7BQ6_K9.6 GBno165details
Yi 6B 200K6.1BQ8_07.0 GBQ8_0142details
Yi 6B6.1BQ8_07.0 GBQ8_0142details
Molmo2 4B4.9BF1610.0 GBQ8_0103details
WeVisOne VL 4B4.8BF1610.0 GBQ8_0103details
Mage VL Cowork4.7BF1610.0 GBQ8_0103details
Qianfan OCR4.7BF1610.1 GBQ8_0103details
Qwen3.5 4B4.7BF169.2 GBF16105details
SingVL S1 4B4.5BF169.0 GBF16108details
LightOnOCR 3 4B4.5BF169.0 GBF16108details
Intern Decision 4B4.5BF169.0 GBF16108details
Qwen3 VL 4B Instruct4.4BF169.3 GBQ8_0113details
Gemma 3 4B IT4.3BF168.4 GBF16117details
MedGemma 1.5 4B IT4.3BF168.4 GBF16117details
MedGemma 4B IT4.3BF168.4 GBF16117details
TranslateGemma 4B IT4.3BF169.2 GBF16106details
Gemma 3 4B PT4.3BF168.4 GBF16117details
Phi 3.5 Vision Instruct4.1BQ8_07.6 GBno207details
Qwen3 4B4.0BF169.3 GBQ8_0113details
Qwen3 4B Instruct 25074.0BF169.3 GBQ8_0113details
Granite Vision 4.1 4B4.0BF167.6 GBF16134details
Phi 4 Mini Instruct3.8BF168.9 GBQ8_0119details
Phi 3 Mini 4K Instruct3.8BQ8_07.2 GBno225details
Phi 3.5 Mini Instruct3.8BQ8_07.2 GBno225details
Phi 3 Mini 128K Instruct3.8BQ8_07.2 GBno225details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF16121details
Granite 4.2 3B3.7BF168.0 GBF16125details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF16125details
Raon 1 Nano3.5BF167.0 GBF16145details
Granite 4.1 3B3.4BF167.6 GBF16134details
Granite 4.0 Micro3.4BF167.6 GBF16134details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16441details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16388details
Tiny Aya Global3.3BF167.5 GBF16136details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16451details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16454details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16454details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16516details
Llama 3.2 3B Instruct3.2BF167.5 GBF16142details
Llama 3.2 3B3.2BF167.5 GBF16142details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF16142details
d1 3B3.1BF165.8 GBF16169details
LFM2.5 VL 3B3.1BF165.8 GBF16169details
Qwen2.5 3B Instruct3.1BF166.7 GBF16148details
SmolLM3 3B3.1BF166.9 GBF16148details
SmolLM3 3B Base3.1BF166.9 GBF16148details
Phi 22.8BF168.2 GBno164details
LFM2.5 2.6B2.7BQ8_03.4 GBQ8_0317details
Granite 3.3 2B Instruct2.5BF165.8 GBF16180details
MiniCPM5 2B2.5BF165.7 GBF16181details
WeVisOne VL 2B2.4BF165.4 GBF16224details
Cosmos Reason2 2B2.4BF165.4 GBF16224details
Qwen3.5 2B2.3BF164.6 GBF16234details
SmolVLM Instruct2.2BF165.4 GBF16252details
Qwen3 1.7B2.0BF165.4 GBF16224details
d1 2B1.9BF164.0 GBF16263details
LFM3 VL 2B1.9BF164.0 GBF16263details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16256details
LFM3 1.7B1.7BF167.2 GBF16132details
LFM3 1.7B Base1.7BF164.0 GBF16263details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16266details
SmolLM2 1.7B1.7BF165.2 GBF16266details
SmolLM 1.7B1.7BF165.2 GBF16266details
Granite 4.0 1B1.6BF164.3 GBF16279details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16295details
OLMo 2 0425 1B1.5BF164.4 GBF16307details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16339details
Llama 3.2 1B1.2BF163.2 GBF16368details
Llama 3.2 1B Instruct1.2BF163.2 GBF16368details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16195details
LFM2 1.2B1.2BF163.0 GBF16389details
HunyuanOCR1.1BF163.5 GBF16407details
PaddleOCR VL 1.6959MF162.5 GBF16476details
Qwen3.5 0.8B873MF162.4 GBF16585details
LightOnOCR 3 0.8B853MF162.4 GBF16601details
FastVLM 0.5B759MF162.2 GBF16601details
Qwen3 0.6B752MF163.0 GBF16604details
GOT OCR2.0716MF162.8 GBF16637details
Nemotron 3.5 Parse631MF161.9 GBF16698details
GOT OCR 2.0 HF561MF162.5 GBF16814details
SmolVLM2 500M Video Instruct507MF161.6 GBF161112details
SmolVLM 500M Instruct507MF161.6 GBF161112details
Qwen2.5 0.5B Instruct494MF161.7 GBF16923details
LFM2.5 VL 450M449MF161.3 GBF161282details
SmolLM2 360M Instruct362MF161.5 GBF161257details
SmolLM2 360M362MF161.5 GBF161257details
LFM2.5 350M354MF161.9 GBF16641details

Bigger models with CPU offload

These popular models don't fit in the RTX 3080 12GB's VRAM at 4-bit, but run with llama.cpp offload, keeping part of the model in system RAM. Speed range is DDR4-3200 to DDR5-6000 (estimate). Method.

ModelParamsOffload plan
GLM 5.3 Flash321B MoEOffload Q2_K, 128 GB RAM · ~7.1–14 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~13–27 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q2_K, 128 GB RAM · ~8.1–15 tok/s
DeepSeek V4 Flash291B MoEOffload Q2_K, 128 GB RAM · ~9.5–19 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 128 GB RAM · ~5.4–11 tok/s
MiniMax M2.5229B MoEOffload Q3_K_M, 128 GB RAM · ~6.9–14 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.1–8.4 tok/s
Nex N2.6321B MoEOffload Q2_K, 128 GB RAM · ~6.8–14 tok/s

Just too big at 4-bit

These need up to 35% more memory than the RTX 3080 12GB has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
gpt-oss Safeguard 20B11.8 GB0.4 GB
gpt-oss 20B11.8 GB0.4 GB
DeepSeek MoE 16B Base12.5 GB1.1 GB
ERNIE 4.5 21B A3B PT13.6 GB2.2 GB
ERNIE 4.5 21B A3B Thinking13.6 GB2.2 GB
LFM2.5 24B A2B14.3 GB2.9 GB

Questions

What is the biggest model that runs on the RTX 3080 12GB?

Of the models we track, the largest that fits at 8K context is GLM 4.7 Flash (31B) at Q2_K, needing about 11.4 GB.

How much of the RTX 3080 12GB's memory can a model use?

About 11.4 GB: we assume 95% of the 12 GB is usable after driver and display overhead.

Spec source: NVIDIA specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.