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

What AI models run on the RX 7700 XT?

The RX 7700 XT has 12 GB of VRAM (432 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
432 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, RTX 5070, Arc B580.

Using a specific tool? Ollama models that run on the RX 7700 XT · 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 GBno265details
Muse Glimmer 30B30BQ2_K11.4 GBno39details
Raon 1 Mini Preview29B (4.3B act.)Q2_K11.2 GBno267details
Arete 27B27BQ2_K11.2 GBno41details
MiniMax H3 Image PE27BQ2_K11.3 GBno40details
Trinity Mini26B (2.0B act.)Q2_K9.8 GBQ2_K609details
GLiDE 27B26BQ2_K10.5 GBno44details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q2_K10.4 GBQ2_K288details
LFM2.5 24B A2B24B (2.0B act.)Q2_K9.2 GBQ2_K576details
ERNIE 4.5 21B A3B PT22BQ2_K8.6 GBQ2_K53details
ERNIE 4.5 21B A3B Thinking22BQ2_K8.6 GBQ2_K53details
Kimi VL A3B Thinking 250616B (3.0B act.)Q3_K_M8.4 GBQ3_K_M295details
Kimi VL A3B Thinking16B (3.0B act.)Q3_K_M8.4 GBQ3_K_M295details
Kimi VL A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M285details
DeepSeek MoE 16B Base16B (2.8B act.)Q3_K_M10.2 GBno295details
LLaDA2.1 Mini16B (1.4B act.)Q3_K_M8.4 GBQ3_K_M617details
Moonlight 16B A3B Instruct16B (3.0B act.)Q3_K_M8.7 GBQ3_K_M277details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q3_K_M8.0 GBQ3_K_M332details
DeepSeek V2 Lite16B (2.7B act.)Q4_K_M9.8 GBQ3_K_M266details
DeepSeek V2 Lite Chat16B (2.7B act.)Q3_K_M8.4 GBQ3_K_M314details
Qwen2.5 Coder 14B Instruct15BQ3_K_M9.0 GBno59details
DeepSeek R1 Distill Qwen 14B15BQ3_K_M9.0 GBno59details
Qwen3 14B15BQ3_K_M8.7 GBQ2_K60details
Phi 415BQ3_K_M9.0 GBno59details
Gemma 3 12B IT12BQ5_K_M9.6 GBQ4_K_M51details
Llama Guard 4 12B12BQ5_K_M10.2 GBQ2_K51details
KORMo Web 11B11BQ5_K_M9.3 GBQ2_K54details
Llama 3.2 11B Vision Instruct11BQ6_K10.0 GBQ2_K49details
Llama 3.2 11B Vision11BQ6_K10.0 GBQ2_K50details
GLM 4.6V Flash10BQ6_K8.7 GBQ6_K52details
GLM 4.1V 9B Thinking10BQ6_K8.7 GBQ6_K52details
Step3 VL 10B10BQ8_09.9 GBQ4_K_M50details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ6_K45details
Qwen3.5 9B9.7BQ8_010.3 GBQ6_K44details
Ornith 1.5 9B9.7BQ8_010.0 GBQ6_K45details
SenseNova Loop 9B9.4BQ8_010.0 GBQ6_K45details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ6_K45details
GLM 4 9B9.4BQ6_K8.7 GBQ6_K52details
Apertus v1.5 8B8.9BQ6_K8.4 GBQ4_K_M59details
InternLM3 8B Instruct8.8BQ8_09.7 GBQ6_K46details
Granite 4.2 8B8.8BQ6_K8.6 GBQ3_K_M60details
Granite 4.1 8B8.8BQ6_K8.6 GBQ3_K_M60details
Granite 4.1 8B Llmon Preview8.8BQ6_K8.6 GBQ3_K_M60details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ4_K_M50details
Molmo2 8B8.7BQ8_09.9 GBQ4_K_M50details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0406details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_053details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_053details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_053details
olmOCR 7B 0225 Preview8.3BQ8_08.7 GBQ8_053details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_053details
Qwen3 8B8.2BQ8_09.9 GBQ4_K_M50details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ4_K_M50details
Granite 3.3 8B Instruct8.2BQ8_09.8 GBQ4_K_M50details
Granite 3.1 8B Instruct8.2BQ8_09.8 GBQ4_K_M50details
Granite 3.2 8B Instruct8.2BQ8_09.8 GBQ4_K_M50details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ5_K_M50details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M51details
Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M51details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ5_K_M51details
Llama Guard 3 8B8.0BQ8_09.6 GBQ5_K_M51details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ5_K_M51details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ5_K_M51details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ5_K_M51details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ5_K_M51details
Aya Expanse 8B8.0BQ8_09.8 GBQ4_K_M51details
Ling 3.0 Tiny7.9B (1.4B act.)Q8_08.7 GBQ8_0293details
MiMo 7B RL7.8BQ8_09.4 GBQ4_K_M53details
InternLM2.5 7B Chat7.7BQ8_09.2 GBQ5_K_M53details
InternLM2 Chat 7B7.7BQ8_09.2 GBQ5_K_M53details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_053details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_053details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_053details
Olmo 3 7B Instruct7.3BQ6_K8.7 GBQ4_K_M72details
Olmo 3 1025 7B7.3BQ6_K8.7 GBQ4_K_M72details
Olmo 3 7B Think7.3BQ6_K8.7 GBQ4_K_M72details
Mistral 7B v0.17.2BQ8_08.6 GBQ6_K56details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ6_K56details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ6_K56details
Falcon 7B7.2BQ8_07.7 GBQ8_056details
Falcon 7B Instruct7.2BQ8_07.4 GBQ8_059details
Granite 4.0 H Tiny6.9B (1.5B act.)Q8_07.5 GBQ8_0277details
OLMoE 1B 7B 09246.9B (1.3B act.)Q8_08.4 GBQ6_K317details
DeepSeek Coder 7B Instruct v1.56.9BQ6_K9.6 GBno76details
DeepSeek Coder 6.7B Instruct6.7BQ6_K9.6 GBno78details
DeepSeek Coder 6.7B Base6.7BQ6_K9.6 GBno78details
Llama 2 7B HF6.7BQ6_K9.6 GBno78details
Yi 6B 200K6.1BQ8_07.0 GBQ8_067details
Yi 6B6.1BQ8_07.0 GBQ8_067details
Molmo2 4B4.9BF1610.0 GBQ8_049details
WeVisOne VL 4B4.8BF1610.0 GBQ8_049details
Mage VL Cowork4.7BF1610.0 GBQ8_049details
Qianfan OCR4.7BF1610.1 GBQ8_049details
Qwen3.5 4B4.7BF169.2 GBF1650details
SingVL S1 4B4.5BF169.0 GBF1651details
LightOnOCR 3 4B4.5BF169.0 GBF1651details
Intern Decision 4B4.5BF169.0 GBF1651details
Qwen3 VL 4B Instruct4.4BF169.3 GBQ8_054details
Gemma 3 4B IT4.3BF168.4 GBF1656details
MedGemma 1.5 4B IT4.3BF168.4 GBF1656details
MedGemma 4B IT4.3BF168.4 GBF1656details
TranslateGemma 4B IT4.3BF169.2 GBF1650details
Gemma 3 4B PT4.3BF168.4 GBF1656details
Phi 3.5 Vision Instruct4.1BQ8_07.6 GBno98details
Qwen3 4B4.0BF169.3 GBQ8_054details
Qwen3 4B Instruct 25074.0BF169.3 GBQ8_054details
Granite Vision 4.1 4B4.0BF167.6 GBF1663details
Phi 4 Mini Instruct3.8BF168.9 GBQ8_056details
Phi 3 Mini 4K Instruct3.8BQ8_07.2 GBno106details
Phi 3.5 Mini Instruct3.8BQ8_07.2 GBno106details
Phi 3 Mini 128K Instruct3.8BQ8_07.2 GBno106details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF1658details
Granite 4.2 3B3.7BF168.0 GBF1659details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF1659details
Raon 1 Nano3.5BF167.0 GBF1669details
Granite 4.1 3B3.4BF167.6 GBF1663details
Granite 4.0 Micro3.4BF167.6 GBF1663details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16209details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16184details
Tiny Aya Global3.3BF167.5 GBF1664details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16214details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16215details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16215details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16244details
Llama 3.2 3B Instruct3.2BF167.5 GBF1667details
Llama 3.2 3B3.2BF167.5 GBF1667details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1667details
d1 3B3.1BF165.8 GBF1680details
LFM2.5 VL 3B3.1BF165.8 GBF1680details
Qwen2.5 3B Instruct3.1BF166.7 GBF1670details
SmolLM3 3B3.1BF166.9 GBF1670details
SmolLM3 3B Base3.1BF166.9 GBF1670details
Phi 22.8BF168.2 GBno78details
LFM2.5 2.6B2.7BQ8_03.4 GBQ8_0150details
Granite 3.3 2B Instruct2.5BF165.8 GBF1685details
MiniCPM5 2B2.5BF165.7 GBF1686details
WeVisOne VL 2B2.4BF165.4 GBF16106details
Cosmos Reason2 2B2.4BF165.4 GBF16106details
Qwen3.5 2B2.3BF164.6 GBF16111details
SmolVLM Instruct2.2BF165.4 GBF16119details
Qwen3 1.7B2.0BF165.4 GBF16106details
d1 2B1.9BF164.0 GBF16125details
LFM3 VL 2B1.9BF164.0 GBF16125details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16121details
LFM3 1.7B1.7BF167.2 GBF1662details
LFM3 1.7B Base1.7BF164.0 GBF16125details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16126details
SmolLM2 1.7B1.7BF165.2 GBF16126details
SmolLM 1.7B1.7BF165.2 GBF16126details
Granite 4.0 1B1.6BF164.3 GBF16132details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16140details
OLMo 2 0425 1B1.5BF164.4 GBF16145details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16160details
Llama 3.2 1B1.2BF163.2 GBF16174details
Llama 3.2 1B Instruct1.2BF163.2 GBF16174details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1692details
LFM2 1.2B1.2BF163.0 GBF16184details
HunyuanOCR1.1BF163.5 GBF16193details
PaddleOCR VL 1.6959MF162.5 GBF16225details
Qwen3.5 0.8B873MF162.4 GBF16277details
LightOnOCR 3 0.8B853MF162.4 GBF16285details
FastVLM 0.5B759MF162.2 GBF16285details
Qwen3 0.6B752MF163.0 GBF16286details
GOT OCR2.0716MF162.8 GBF16302details
Nemotron 3.5 Parse631MF161.9 GBF16331details
GOT OCR 2.0 HF561MF162.5 GBF16385details
SmolVLM2 500M Video Instruct507MF161.6 GBF16527details
SmolVLM 500M Instruct507MF161.6 GBF16527details
Qwen2.5 0.5B Instruct494MF161.7 GBF16437details
LFM2.5 VL 450M449MF161.3 GBF16607details
SmolLM2 360M Instruct362MF161.5 GBF16595details
SmolLM2 360M362MF161.5 GBF16595details
LFM2.5 350M354MF161.9 GBF16304details

Bigger models with CPU offload

These popular models don't fit in the RX 7700 XT'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 · ~6.8–13 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~13–25 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q2_K, 128 GB RAM · ~7.3–13 tok/s
DeepSeek V4 Flash291B MoEOffload Q2_K, 128 GB RAM · ~9.0–17 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 128 GB RAM · ~5.1–9.6 tok/s
MiniMax M2.5229B MoEOffload Q3_K_M, 128 GB RAM · ~6.7–13 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.0–7.9 tok/s

Comparing cards? The RX 7700 XT (launch MSRP $449) 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 RX 7700 XT 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 RX 7700 XT?

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 RX 7700 XT'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: AMD specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.