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

What AI models run on the RTX 3070?

The RTX 3070 has 8 GB of VRAM (448 GB/s). About 7.6 GB is usable for a model. 161 of 298 open models we track fit at 8K context.

161models fit (8K ctx)
114still fit at 32K ctx
16Blargest that fits
448 GB/smemory bandwidth

This card has 8 GB of VRAM. See every model that fits in 8 GB VRAM. Other 8 GB hardware: RTX 3060 Ti, RTX 4060, RTX 4060 Ti 8GB, RTX 5060, RX 7600.

Using a specific tool? Ollama models that run on the RTX 3070 · LM Studio · llama.cpp offload

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)Details
Kimi VL A3B Thinking 250616B (3.0B act.)Q2_K6.7 GBQ2_K398details
Kimi VL A3B Thinking16B (3.0B act.)Q2_K6.7 GBQ2_K398details
Kimi VL A3B Instruct16B (3.0B act.)Q2_K7.1 GBno372details
LLaDA2.1 Mini16B (1.4B act.)Q2_K6.7 GBno834details
Moonlight 16B A3B Instruct16B (3.0B act.)Q2_K7.1 GBno362details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q2_K6.3 GBQ2_K449details
DeepSeek V2 Lite16B (2.7B act.)Q2_K6.3 GBQ2_K449details
DeepSeek V2 Lite Chat16B (2.7B act.)Q2_K6.8 GBQ2_K411details
Qwen2.5 Coder 14B Instruct15BQ2_K7.6 GBno78details
DeepSeek R1 Distill Qwen 14B15BQ2_K7.6 GBno78details
Qwen3 14B15BQ2_K7.1 GBno81details
Phi 415BQ2_K7.3 GBno81details
Gemma 3 12B IT12BQ2_K6.2 GBno94details
Llama Guard 4 12B12BQ2_K6.5 GBno100details
KORMo Web 11B11BQ2_K5.8 GBno107details
Llama 3.2 11B Vision Instruct11BQ3_K_M6.8 GBno86details
Llama 3.2 11B Vision11BQ3_K_M6.7 GBno86details
GLM 4.6V Flash10BQ4_K_M6.7 GBQ3_K_M73details
GLM 4.1V 9B Thinking10BQ4_K_M6.7 GBQ3_K_M73details
Step3 VL 10B10BQ4_K_M6.5 GBno89details
SenseNova Loop 9B Preview9.7BQ4_K_M6.4 GBQ3_K_M80details
Qwen3.5 9B9.7BQ3_K_M5.9 GBQ3_K_M87details
Ornith 1.5 9B9.7BQ4_K_M6.6 GBQ3_K_M76details
SenseNova Loop 9B9.4BQ4_K_M6.4 GBQ3_K_M80details
MiMo V2.6 Distill Qwen 9B9.4BQ4_K_M6.6 GBQ3_K_M77details
GLM 4 9B9.4BQ4_K_M6.8 GBQ3_K_M72details
Apertus v1.5 8B8.9BQ4_K_M6.7 GBno82details
InternLM3 8B Instruct8.8BQ4_K_M6.0 GBQ3_K_M84details
Granite 4.2 8B8.8BQ4_K_M6.8 GBno84details
Granite 4.1 8B8.8BQ4_K_M6.8 GBno84details
Granite 4.1 8B Llmon Preview8.8BQ4_K_M6.8 GBno84details
Qwen3 VL 8B Instruct8.8BQ4_K_M6.5 GBno89details
Molmo2 8B8.7BQ4_K_M6.5 GBno89details
LFM2.5 8B A1B8.5B (1.0B act.)Q5_K_M6.4 GBQ5_K_M629details
Qwen2.5 VL 7B Instruct8.3BQ5_K_M6.2 GBQ4_K_M82details
UI TARS 1.5 7B8.3BQ5_K_M6.2 GBQ4_K_M82details
olmOCR 2 7B 10258.3BQ5_K_M6.2 GBQ4_K_M82details
olmOCR 7B 0225 Preview8.3BQ5_K_M6.2 GBQ4_K_M82details
UI TARS 7B DPO8.3BQ5_K_M6.2 GBQ4_K_M82details
Qwen3 8B8.2BQ4_K_M6.5 GBno89details
DeepSeek R1 0528 Qwen3 8B8.2BQ4_K_M6.5 GBno89details
Granite 3.3 8B Instruct8.2BQ4_K_M6.3 GBno91details
Granite 3.1 8B Instruct8.2BQ4_K_M6.3 GBno91details
Granite 3.2 8B Instruct8.2BQ4_K_M6.3 GBno91details
Apertus 8B Instruct 25098.1BQ4_K_M6.4 GBno89details
Llama 3.1 8B Instruct8.0BQ4_K_M6.2 GBno91details
Llama 3.1 8B8.0BQ4_K_M6.2 GBno91details
Hermes 3 Llama 3.1 8B8.0BQ4_K_M6.2 GBno91details
Llama Guard 3 8B8.0BQ4_K_M6.2 GBno91details
DeepSeek R1 Distill Llama 8B8.0BQ4_K_M6.2 GBno91details
Meta Llama 3 8B Instruct8.0BQ4_K_M6.2 GBno91details
Meta Llama Guard 2 8B8.0BQ4_K_M6.2 GBno91details
Meta Llama 3.1 8B Instruct8.0BQ4_K_M6.2 GBno91details
Aya Expanse 8B8.0BQ4_K_M6.6 GBno89details
Ling 3.0 Tiny7.9B (1.4B act.)Q5_K_M6.2 GBQ4_K_M448details
MiMo 7B RL7.8BQ4_K_M6.2 GBno96details
InternLM2.5 7B Chat7.7BQ5_K_M6.7 GBQ2_K81details
InternLM2 Chat 7B7.7BQ5_K_M6.7 GBQ2_K81details
Qwen2.5 7B Instruct7.6BQ5_K_M6.2 GBQ4_K_M82details
DeepSeek R1 Distill Qwen 7B7.6BQ5_K_M6.2 GBQ4_K_M82details
Qwen2.5 Coder 7B Instruct7.6BQ5_K_M6.2 GBQ4_K_M82details
Olmo 3 7B Instruct7.3BQ3_K_M6.5 GBno123details
Olmo 3 1025 7B7.3BQ3_K_M6.4 GBno126details
Olmo 3 7B Think7.3BQ3_K_M6.5 GBno123details
Mistral 7B v0.17.2BQ5_K_M6.2 GBQ2_K87details
Mistral 7B Instruct v0.27.2BQ5_K_M6.2 GBQ2_K87details
Mistral 7B Instruct v0.17.2BQ5_K_M6.2 GBQ2_K87details
Falcon 7B7.2BQ5_K_M5.9 GBQ5_K_M78details
Falcon 7B Instruct7.2BQ5_K_M5.7 GBQ5_K_M81details
Granite 4.0 H Tiny6.9B (1.5B act.)Q6_K6.0 GBQ6_K371details
OLMoE 1B 7B 09246.9B (1.3B act.)Q6_K6.8 GBQ2_K425details
DeepSeek Coder 7B Instruct v1.56.9BQ3_K_M7.6 GBno129details
DeepSeek Coder 6.7B Instruct6.7BQ3_K_M7.5 GBno136details
DeepSeek Coder 6.7B Base6.7BQ3_K_M7.5 GBno136details
Llama 2 7B HF6.7BQ2_K6.8 GBno177details
Yi 6B 200K6.1BQ6_K5.7 GBQ5_K_M90details
Yi 6B6.1BQ6_K5.7 GBQ5_K_M90details
Molmo2 4B4.9BQ8_06.2 GBQ3_K_M95details
WeVisOne VL 4B4.8BQ8_06.2 GBQ3_K_M95details
Mage VL Cowork4.7BQ8_06.2 GBQ3_K_M95details
Qianfan OCR4.7BQ8_06.2 GBQ3_K_M95details
Qwen3.5 4B4.7BQ8_05.4 GBQ8_097details
SingVL S1 4B4.5BQ8_05.3 GBQ8_0100details
LightOnOCR 3 4B4.5BQ8_05.3 GBQ8_0100details
Intern Decision 4B4.5BQ8_05.3 GBQ8_0100details
Qwen3 VL 4B Instruct4.4BQ8_05.8 GBQ2_K105details
Gemma 3 4B IT4.3BQ8_05.0 GBQ8_0108details
MedGemma 1.5 4B IT4.3BQ8_05.0 GBQ8_0108details
MedGemma 4B IT4.3BQ8_05.0 GBQ8_0108details
TranslateGemma 4B IT4.3BQ8_05.0 GBQ8_0108details
Gemma 3 4B PT4.3BQ8_05.0 GBQ8_0108details
Phi 3.5 Vision Instruct4.1BQ6_K6.6 GBno132details
Qwen3 4B4.0BQ8_05.8 GBQ2_K105details
Qwen3 4B Instruct 25074.0BQ8_05.8 GBQ2_K105details
Granite Vision 4.1 4B4.0BQ8_04.6 GBQ8_0124details
Phi 4 Mini Instruct3.8BQ8_05.6 GBQ3_K_M110details
Phi 3 Mini 4K Instruct3.8BQ6_K6.4 GBno143details
Phi 3.5 Mini Instruct3.8BQ6_K6.4 GBno143details
Phi 3 Mini 128K Instruct3.8BQ6_K6.4 GBno143details
Qwen2.5 VL 3B Instruct3.8BQ8_04.0 GBQ8_0136details
Granite 4.2 3B3.7BQ8_04.8 GBQ8_0115details
Granite 4.1 3B Llmon Preview3.7BQ8_04.8 GBQ8_0115details
Raon 1 Nano3.5BQ8_04.3 GBQ8_0132details
Granite 4.1 3B3.4BQ8_04.6 GBQ8_0124details
Granite 4.0 Micro3.4BQ8_04.6 GBQ8_0124details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBQ8_0216details
DeepSeek VL2 Tiny3.4B (1.2B act.)Q8_04.5 GBQ8_0359details
Tiny Aya Global3.3BQ8_04.6 GBQ8_0125details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBQ8_0222details
Unlimited OCR3.3B (1.1B act.)F166.6 GBQ8_0223details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBQ8_0223details
Granite 3.1 3B A800M Instruct3.3B (883M act.)Q8_04.3 GBQ8_0477details
Llama 3.2 3B Instruct3.2BQ8_04.7 GBQ6_K131details
Llama 3.2 3B3.2BQ8_04.7 GBQ6_K131details
Hermes 3 Llama 3.2 3B3.2BQ8_04.7 GBQ6_K131details
d1 3B3.1BF165.8 GBF1683details
LFM2.5 VL 3B3.1BF165.8 GBF1683details
Qwen2.5 3B Instruct3.1BF166.7 GBQ8_073details
SmolLM3 3B3.1BQ8_04.3 GBQ8_0137details
SmolLM3 3B Base3.1BQ8_04.3 GBQ8_0137details
Phi 22.8BQ8_05.8 GBno151details
LFM2.5 2.6B2.7BQ8_03.4 GBQ8_0156details
Granite 3.3 2B Instruct2.5BF165.8 GBQ8_088details
MiniCPM5 2B2.5BF165.7 GBF1689details
WeVisOne VL 2B2.4BF165.4 GBQ8_0110details
Cosmos Reason2 2B2.4BF165.4 GBQ8_0110details
Qwen3.5 2B2.3BF164.6 GBF16115details
SmolVLM Instruct2.2BF165.4 GBQ4_K_M124details
Qwen3 1.7B2.0BF165.4 GBQ8_0110details
d1 2B1.9BF164.0 GBF16129details
LFM3 VL 2B1.9BF164.0 GBF16129details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16126details
LFM3 1.7B1.7BQ8_02.5 GBQ8_0243details
LFM3 1.7B Base1.7BF164.0 GBF16129details
SmolLM2 1.7B Instruct1.7BF165.2 GBQ4_K_M131details
SmolLM2 1.7B1.7BF165.2 GBQ4_K_M131details
SmolLM 1.7B1.7BF165.2 GBQ4_K_M131details
Granite 4.0 1B1.6BF164.3 GBF16137details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16145details
OLMo 2 0425 1B1.5BF164.4 GBQ8_0151details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBQ6_K166details
Llama 3.2 1B1.2BF163.2 GBF16181details
Llama 3.2 1B Instruct1.2BF163.2 GBF16181details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1696details
LFM2 1.2B1.2BF163.0 GBF16191details
HunyuanOCR1.1BF163.5 GBF16200details
PaddleOCR VL 1.6959MF162.5 GBF16234details
Qwen3.5 0.8B873MF162.4 GBF16288details
LightOnOCR 3 0.8B853MF162.4 GBF16295details
FastVLM 0.5B759MF162.2 GBF16295details
Qwen3 0.6B752MF163.0 GBF16297details
GOT OCR2.0716MF162.8 GBF16313details
Nemotron 3.5 Parse631MF161.9 GBF16343details
GOT OCR 2.0 HF561MF162.5 GBF16400details
SmolVLM2 500M Video Instruct507MF161.6 GBF16546details
SmolVLM 500M Instruct507MF161.6 GBF16546details
Qwen2.5 0.5B Instruct494MF161.7 GBF16453details
LFM2.5 VL 450M449MF161.3 GBF16630details
SmolLM2 360M Instruct362MF161.5 GBF16617details
SmolLM2 360M362MF161.5 GBF16617details
LFM2.5 350M354MF161.9 GBF16315details

Bigger models with CPU offload

These popular models don't fit in the RTX 3070'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
Qwen3.8 27B28BOffload Q4_K_M, 32 GB RAM · ~2.0–4.0 tok/s
GLM 5.3 Flash321B MoEOffload Q2_K, 128 GB RAM · ~6.7–13 tok/s
Gemma 4 31B IT31BOffload Q4_K_M, 32 GB RAM · ~1.6–3.3 tok/s
gpt-oss 20B21B MoEOffload Q4_K_M, 32 GB RAM · ~29–52 tok/s
Gemma 4 26B A4B IT26B MoEOffload Q4_K_M, 32 GB RAM · ~27–47 tok/s
Qwen3.6 35B A3B36B MoEOffload Q4_K_M, 32 GB RAM · ~34–55 tok/s
Qwen3 32B33BOffload Q4_K_M, 32 GB RAM · ~1.6–3.3 tok/s
Qwen3.6 27B28BOffload Q4_K_M, 32 GB RAM · ~1.9–3.8 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~12–24 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q2_K, 128 GB RAM · ~2.3–4.7 tok/s
Qwen3 Coder 30B A3B Instruct31B MoEOffload Q4_K_M, 32 GB RAM · ~26–48 tok/s
Qwen3.5 27B28BOffload Q4_K_M, 32 GB RAM · ~1.9–3.8 tok/s
Qwen3.5 35B A3B36B MoEOffload Q4_K_M, 32 GB RAM · ~34–55 tok/s
DeepSeek V4 Flash291B MoEOffload Q2_K, 128 GB RAM · ~8.7–17 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 128 GB RAM · ~2.5–5.2 tok/s
MiniMax M2.5229B MoEOffload Q3_K_M, 128 GB RAM · ~6.5–13 tok/s
Hy3299B MoEOffload Q2_K, 128 GB RAM · ~5.1–10 tok/s
Nex N2.6321B MoEOffload Q2_K, 128 GB RAM · ~6.4–12 tok/s

Comparing cards? The RTX 3070 (launch MSRP $499) is in best GPUs for local LLMs under $500. See also Mac vs PC.

Just too big at 4-bit

These need up to 35% more memory than the RTX 3070 has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
Llama 3.2 11B Vision8.0 GB0.4 GB
KORMo Web 11B8.3 GB0.7 GB
DeepSeek Coder 6.7B Instruct8.3 GB0.7 GB
DeepSeek Coder 6.7B Base8.3 GB0.7 GB
DeepSeek Coder 7B Instruct v1.58.3 GB0.7 GB
Llama 2 7B HF8.3 GB0.7 GB
Gemma 3 12B IT8.5 GB0.9 GB
Llama Guard 4 12B9.1 GB1.5 GB
DeepSeek V2 Lite9.8 GB2.2 GB

Questions

What is the biggest model that runs on the RTX 3070?

Of the models we track, the largest that fits at 8K context is Kimi VL A3B Thinking 2506 (16B) at Q2_K, needing about 6.7 GB.

How much of the RTX 3070's memory can a model use?

About 7.6 GB: we assume 95% of the 8 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.