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

What AI models run on the RTX 5070?

The RTX 5070 has 12 GB of VRAM (672 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
672 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 3080 12GB, RTX 4070, RTX 4070 Super, RTX 4070 Ti, RX 7700 XT, Arc B580.

Compare for local LLMs: RTX 5070 vs RTX 5060 Ti 16GBRTX 5070 vs RTX 4070RTX 5070 vs RX 9070 all comparisons

Using a specific tool? Ollama models that run on the RTX 5070 · 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 GBno412details
Muse Glimmer 30B30BQ2_K11.4 GBno60details
Raon 1 Mini Preview29B (4.3B act.)Q2_K11.2 GBno415details
Arete 27B27BQ2_K11.2 GBno64details
MiniMax H3 Image PE27BQ2_K11.3 GBno63details
Trinity Mini26B (2.0B act.)Q2_K9.8 GBQ2_K948details
GLiDE 27B26BQ2_K10.5 GBno69details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q2_K10.4 GBQ2_K448details
LFM2.5 24B A2B24B (2.0B act.)Q2_K9.2 GBQ2_K896details
ERNIE 4.5 21B A3B PT22BQ2_K8.6 GBQ2_K83details
ERNIE 4.5 21B A3B Thinking22BQ2_K8.6 GBQ2_K83details
Kimi VL A3B Thinking 250616B (3.0B act.)Q3_K_M8.4 GBQ3_K_M458details
Kimi VL A3B Thinking16B (3.0B act.)Q3_K_M8.4 GBQ3_K_M458details
Kimi VL A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M443details
DeepSeek MoE 16B Base16B (2.8B act.)Q3_K_M10.2 GBno459details
LLaDA2.1 Mini16B (1.4B act.)Q3_K_M8.4 GBQ3_K_M959details
Moonlight 16B A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M431details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q3_K_M8.0 GBQ3_K_M517details
DeepSeek V2 Lite16B (2.7B act.)Q4_K_M9.8 GBQ3_K_M413details
DeepSeek V2 Lite Chat16B (2.7B act.)Q3_K_M8.4 GBQ3_K_M488details
Qwen2.5 Coder 14B Instruct15BQ3_K_M9.0 GBno92details
DeepSeek R1 Distill Qwen 14B15BQ3_K_M9.0 GBno92details
Qwen3 14B15BQ3_K_M8.7 GBQ2_K93details
Phi 415BQ3_K_M9.0 GBno91details
Gemma 3 12B IT12BQ5_K_M9.6 GBQ4_K_M80details
Llama Guard 4 12B12BQ5_K_M10.2 GBQ2_K79details
KORMo Web 11B11BQ5_K_M9.3 GBQ2_K84details
Llama 3.2 11B Vision Instruct11BQ6_K10.0 GBQ2_K77details
Llama 3.2 11B Vision11BQ6_K10.0 GBQ2_K77details
GLM 4.6V Flash10BQ6_K8.7 GBQ6_K81details
GLM 4.1V 9B Thinking10BQ6_K8.7 GBQ6_K81details
Step3 VL 10B10BQ8_09.9 GBQ4_K_M77details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ6_K70details
Qwen3.5 9B9.7BQ8_010.3 GBQ6_K69details
Ornith 1.5 9B9.7BQ8_010.0 GBQ6_K70details
SenseNova Loop 9B9.4BQ8_010.0 GBQ6_K70details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ6_K70details
GLM 4 9B9.4BQ6_K8.7 GBQ6_K81details
Apertus v1.5 8B8.9BQ6_K8.4 GBQ4_K_M92details
InternLM3 8B Instruct8.8BQ8_09.7 GBQ6_K72details
Granite 4.2 8B8.8BQ6_K8.6 GBQ3_K_M93details
Granite 4.1 8B8.8BQ6_K8.6 GBQ3_K_M93details
Granite 4.1 8B Llmon Preview8.8BQ6_K8.6 GBQ3_K_M93details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ4_K_M77details
Molmo2 8B8.7BQ8_09.9 GBQ4_K_M77details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0632details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_083details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_083details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_083details
olmOCR 7B 0225 Preview8.3BQ8_08.7 GBQ8_083details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_083details
Qwen3 8B8.2BQ8_09.9 GBQ4_K_M77details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ4_K_M77details
Granite 3.3 8B Instruct8.2BQ8_09.8 GBQ4_K_M77details
Granite 3.1 8B Instruct8.2BQ8_09.8 GBQ4_K_M77details
Granite 3.2 8B Instruct8.2BQ8_09.8 GBQ4_K_M77details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ5_K_M78details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M79details
Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M79details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M79details
Llama Guard 3 8B8.0BQ8_09.6 GBQ5_K_M79details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ5_K_M79details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ5_K_M79details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ5_K_M79details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M79details
Aya Expanse 8B8.0BQ8_09.8 GBQ4_K_M79details
Ling 3.0 Tiny7.9B (1.4B act.)Q8_08.7 GBQ8_0456details
MiMo 7B RL7.8BQ8_09.4 GBQ4_K_M83details
InternLM2.5 7B Chat7.7BQ8_09.2 GBQ5_K_M82details
InternLM2 Chat 7B7.7BQ8_09.2 GBQ5_K_M82details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_083details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_083details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_083details
Olmo 3 7B Instruct7.3BQ6_K8.7 GBQ4_K_M112details
Olmo 3 1025 7B7.3BQ6_K8.7 GBQ4_K_M112details
Olmo 3 7B Think7.3BQ6_K8.7 GBQ4_K_M112details
Mistral 7B v0.17.2BQ8_08.6 GBQ6_K87details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ6_K87details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ6_K87details
Falcon 7B7.2BQ8_07.7 GBQ8_088details
Falcon 7B Instruct7.2BQ8_07.4 GBQ8_091details
Granite 4.0 H Tiny6.9B (1.5B act.)Q8_07.5 GBQ8_0431details
OLMoE 1B 7B 09246.9B (1.3B act.)Q8_08.4 GBQ6_K493details
DeepSeek Coder 7B Instruct v1.56.9BQ6_K9.6 GBno118details
DeepSeek Coder 6.7B Instruct6.7BQ6_K9.6 GBno121details
DeepSeek Coder 6.7B Base6.7BQ6_K9.6 GBno121details
Llama 2 7B HF6.7BQ6_K9.6 GBno122details
Yi 6B 200K6.1BQ8_07.0 GBQ8_0104details
Yi 6B6.1BQ8_07.0 GBQ8_0104details
Molmo2 4B4.9BF1610.0 GBQ8_076details
WeVisOne VL 4B4.8BF1610.0 GBQ8_076details
Mage VL Cowork4.7BF1610.0 GBQ8_076details
Qianfan OCR4.7BF1610.1 GBQ8_076details
Qwen3.5 4B4.7BF169.2 GBF1678details
SingVL S1 4B4.5BF169.0 GBF1680details
LightOnOCR 3 4B4.5BF169.0 GBF1680details
Intern Decision 4B4.5BF169.0 GBF1680details
Qwen3 VL 4B Instruct4.4BF169.3 GBQ8_083details
Gemma 3 4B IT4.3BF168.4 GBF1687details
MedGemma 1.5 4B IT4.3BF168.4 GBF1687details
MedGemma 4B IT4.3BF168.4 GBF1687details
TranslateGemma 4B IT4.3BF169.2 GBF1678details
Gemma 3 4B PT4.3BF168.4 GBF1687details
Phi 3.5 Vision Instruct4.1BQ8_07.6 GBno153details
Qwen3 4B4.0BF169.3 GBQ8_084details
Qwen3 4B Instruct 25074.0BF169.3 GBQ8_083details
Granite Vision 4.1 4B4.0BF167.6 GBF1699details
Phi 4 Mini Instruct3.8BF168.9 GBQ8_088details
Phi 3 Mini 4K Instruct3.8BQ8_07.2 GBno165details
Phi 3.5 Mini Instruct3.8BQ8_07.2 GBno165details
Phi 3 Mini 128K Instruct3.8BQ8_07.2 GBno165details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF1689details
Granite 4.2 3B3.7BF168.0 GBF1692details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF1692details
Raon 1 Nano3.5BF167.0 GBF16107details
Granite 4.1 3B3.4BF167.6 GBF1699details
Granite 4.0 Micro3.4BF167.6 GBF1699details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16325details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16286details
Tiny Aya Global3.3BF167.5 GBF16100details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16332details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16334details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16334details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16380details
Llama 3.2 3B Instruct3.2BF167.5 GBF16104details
Llama 3.2 3B3.2BF167.5 GBF16105details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF16104details
d1 3B3.1BF165.8 GBF16124details
LFM2.5 VL 3B3.1BF165.8 GBF16124details
Qwen2.5 3B Instruct3.1BF166.7 GBF16109details
SmolLM3 3B3.1BF166.9 GBF16109details
SmolLM3 3B Base3.1BF166.9 GBF16109details
Phi 22.8BF168.2 GBno121details
LFM2.5 2.6B2.7BQ8_03.4 GBQ8_0234details
Granite 3.3 2B Instruct2.5BF165.8 GBF16133details
MiniCPM5 2B2.5BF165.7 GBF16133details
WeVisOne VL 2B2.4BF165.4 GBF16165details
Cosmos Reason2 2B2.4BF165.4 GBF16165details
Qwen3.5 2B2.3BF164.6 GBF16172details
SmolVLM Instruct2.2BF165.4 GBF16185details
Qwen3 1.7B2.0BF165.4 GBF16165details
d1 2B1.9BF164.0 GBF16194details
LFM3 VL 2B1.9BF164.0 GBF16194details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16189details
LFM3 1.7B1.7BF167.2 GBF1697details
LFM3 1.7B Base1.7BF164.0 GBF16194details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16196details
SmolLM2 1.7B1.7BF165.2 GBF16196details
SmolLM 1.7B1.7BF165.2 GBF16196details
Granite 4.0 1B1.6BF164.3 GBF16206details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16218details
OLMo 2 0425 1B1.5BF164.4 GBF16226details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16250details
Llama 3.2 1B1.2BF163.2 GBF16271details
Llama 3.2 1B Instruct1.2BF163.2 GBF16271details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16143details
LFM2 1.2B1.2BF163.0 GBF16287details
HunyuanOCR1.1BF163.5 GBF16300details
PaddleOCR VL 1.6959MF162.5 GBF16351details
Qwen3.5 0.8B873MF162.4 GBF16431details
LightOnOCR 3 0.8B853MF162.4 GBF16443details
FastVLM 0.5B759MF162.2 GBF16443details
Qwen3 0.6B752MF163.0 GBF16445details
GOT OCR2.0716MF162.8 GBF16469details
Nemotron 3.5 Parse631MF161.9 GBF16514details
GOT OCR 2.0 HF561MF162.5 GBF16599details
SmolVLM2 500M Video Instruct507MF161.6 GBF16819details
SmolVLM 500M Instruct507MF161.6 GBF16819details
Qwen2.5 0.5B Instruct494MF161.7 GBF16680details
LFM2.5 VL 450M449MF161.3 GBF16945details
SmolLM2 360M Instruct362MF161.5 GBF16926details
SmolLM2 360M362MF161.5 GBF16926details
LFM2.5 350M354MF161.9 GBF16472details

Bigger models with CPU offload

These popular models don't fit in the RTX 5070'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 · ~7.0–14 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~13–26 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q2_K, 128 GB RAM · ~7.8–14 tok/s
DeepSeek V4 Flash291B MoEOffload Q2_K, 128 GB RAM · ~9.3–18 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 128 GB RAM · ~5.3–10 tok/s
MiniMax M2.5229B MoEOffload Q3_K_M, 128 GB RAM · ~6.8–14 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.1–8.2 tok/s

Comparing cards? The RTX 5070 (launch MSRP $549) is in best GPUs for local LLMs under $800. See also Mac vs PC.

Just too big at 4-bit

These need up to 35% more memory than the RTX 5070 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 5070?

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 5070'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.