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

What AI models run on the RX 9070 XT?

The RX 9070 XT has 16 GB of VRAM (640 GB/s). About 15.2 GB is usable for a model. 200 of 298 open models we track fit at 8K context.

200models fit (8K ctx)
189still fit at 32K ctx
37Blargest that fits
640 GB/smemory bandwidth

This card has 16 GB of VRAM. See every model that fits in 16 GB VRAM. Other 16 GB hardware: RTX 4060 Ti 16GB, RTX 4070 Ti Super, RTX 4080, RTX 4080 Super, RTX 5060 Ti 16GB, RTX 5070 Ti, RTX 5080, RX 7800 XT.

Compare for local LLMs: RX 9070 XT vs RTX 5070 Ti all comparisons

Using a specific tool? Ollama models that run on the RX 9070 XT · LM Studio · llama.cpp offload

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)Details
Maxwell V1 Flash37B (2.3B act.)Q2_K14.0 GBQ2_K752details
Qwen3.6 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K568details
Ornith 1.5 35B A3B36B (3.0B act.)Q2_K13.2 GBQ2_K586details
Qwen3.5 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K568details
Sage A1.536B (4.7B act.)Q2_K13.4 GBQ2_K366details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q2_K13.1 GBQ2_K569details
Yi 1.5 34B Chat34BQ2_K14.3 GBno50details
Nous Hermes 2 Yi 34B34BQ2_K14.8 GBno48details
DeepSeek R1 Distill Qwen 32B33BQ2_K14.2 GBno52details
Qwen3 32B33BQ2_K14.1 GBno52details
GLM Z1 32B 041433BQ2_K12.5 GBQ2_K52details
GLM 4 32B 041433BQ2_K12.6 GBQ2_K52details
Olmo 3 32B Think32BQ2_K13.0 GBQ2_K53details
Granite 4.0 H Small32B (8.8B act.)Q2_K11.7 GBQ2_K199details
Gemma 4 31B IT31BQ2_K14.7 GBno51details
Gemma 4 31B31BQ2_K13.9 GBno55details
GLM 4.7 Flash31B (4.6B act.)Q2_K11.4 GBQ2_K393details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q2_K12.1 GBQ2_K569details
North Mini Code 1.030B (3.3B act.)Q2_K11.7 GBQ2_K532details
Hy Memory 30B A3B30B (3.0B act.)Q2_K11.9 GBQ2_K569details
Muse Glimmer 30B30BQ2_K11.4 GBQ2_K57details
Granite 4.2 30B29BQ2_K12.7 GBno59details
Raon 1 Mini Preview29B (4.3B act.)Q2_K11.2 GBQ2_K395details
Qwen3.8 27B28BQ2_K11.5 GBQ2_K59details
Qwen3.6 27B28BQ2_K12.6 GBQ2_K53details
Qwen3.5 27B28BQ2_K12.6 GBQ2_K53details
Gemma 3 27B IT27BQ2_K11.7 GBQ2_K61details
Arete 27B27BQ3_K_M13.6 GBQ2_K49details
MiniMax H3 Image PE27BQ2_K11.3 GBQ2_K60details
Trinity Mini26B (2.0B act.)Q3_K_M12.3 GBQ3_K_M703details
GLiDE 27B26BQ3_K_M13.2 GBQ2_K50details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q3_K_M13.2 GBQ2_K327details
Gemma 4 26B A4B IT26B (4.0B act.)Q3_K_M13.5 GBQ2_K317details
LFM2.5 24B A2B24B (2.0B act.)Q3_K_M11.7 GBQ3_K_M655details
ERNIE 4.5 21B A3B PT22BQ4_K_M13.6 GBQ3_K_M47details
ERNIE 4.5 21B A3B Thinking22BQ4_K_M13.6 GBQ3_K_M47details
gpt-oss Safeguard 20B22B (4.8B act.)Q8_012.3 GBQ8_0237details
gpt-oss 20B21B (4.2B act.)Q8_012.3 GBQ8_0264details
Kimi VL A3B Thinking 250616B (3.0B act.)Q6_K13.5 GBQ5_K_M260details
Kimi VL A3B Thinking16B (3.0B act.)Q6_K13.5 GBQ5_K_M260details
Kimi VL A3B Instruct16B (3.0B act.)Q5_K_M12.2 GBQ5_K_M291details
DeepSeek MoE 16B Base16B (2.8B act.)Q4_K_M12.5 GBQ2_K341details
LLaDA2.1 Mini16B (1.4B act.)Q6_K13.4 GBQ5_K_M545details
Moonlight 16B A3B Instruct16B (3.0B act.)Q5_K_M12.2 GBQ5_K_M283details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q5_K_M11.9 GBQ5_K_M319details
DeepSeek V2 Lite16B (2.7B act.)Q6_K12.8 GBQ6_K293details
DeepSeek V2 Lite Chat16B (2.7B act.)Q5_K_M11.9 GBQ5_K_M319details
Qwen2.5 Coder 14B Instruct15BQ6_K13.5 GBQ3_K_M53details
DeepSeek R1 Distill Qwen 14B15BQ6_K13.5 GBQ3_K_M53details
Qwen3 14B15BQ6_K13.2 GBQ3_K_M53details
Phi 415BQ6_K13.4 GBQ2_K53details
Gemma 3 12B IT12BQ8_013.4 GBQ6_K51details
Llama Guard 4 12B12BQ6_K11.5 GBQ4_K_M65details
KORMo Web 11B11BQ8_013.0 GBQ5_K_M54details
Llama 3.2 11B Vision Instruct11BQ8_011.6 GBQ5_K_M62details
Llama 3.2 11B Vision11BQ8_012.4 GBQ5_K_M57details
GLM 4.6V Flash10BQ8_010.3 GBQ8_064details
GLM 4.1V 9B Thinking10BQ8_010.3 GBQ8_064details
Step3 VL 10B10BQ8_09.9 GBQ8_073details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ8_067details
Qwen3.5 9B9.7BQ8_010.3 GBQ8_065details
Ornith 1.5 9B9.7BQ8_010.0 GBQ8_067details
SenseNova Loop 9B9.4BQ8_010.0 GBQ8_067details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ8_067details
GLM 4 9B9.4BQ8_010.3 GBQ8_064details
Apertus v1.5 8B8.9BQ8_010.5 GBQ8_068details
InternLM3 8B Instruct8.8BQ8_09.7 GBQ8_068details
Granite 4.2 8B8.8BQ8_010.5 GBQ6_K68details
Granite 4.1 8B8.8BQ8_010.5 GBQ6_K68details
Granite 4.1 8B Llmon Preview8.8BQ8_010.5 GBQ6_K68details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ8_073details
Molmo2 8B8.7BQ8_09.9 GBQ8_073details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0601details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_079details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_079details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_079details
olmOCR 7B 0225 Preview8.3BQ8_08.7 GBQ8_079details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_079details
Qwen3 8B8.2BQ8_09.9 GBQ8_073details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ8_073details
Granite 3.3 8B Instruct8.2BQ8_09.8 GBQ8_074details
Granite 3.1 8B Instruct8.2BQ8_09.8 GBQ8_074details
Granite 3.2 8B Instruct8.2BQ8_09.8 GBQ8_074details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ8_075details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_075details
Llama 3.1 8B8.0BQ8_09.6 GBQ8_075details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ8_075details
Llama Guard 3 8B8.0BQ8_09.6 GBQ8_075details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ8_075details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ8_075details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ8_075details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_075details
Aya Expanse 8B8.0BQ8_09.8 GBQ8_075details
Ling 3.0 Tiny7.9B (1.4B act.)Q8_08.7 GBQ8_0435details
MiMo 7B RL7.8BQ8_09.4 GBQ8_079details
InternLM2.5 7B Chat7.7BQ8_09.2 GBQ8_078details
InternLM2 Chat 7B7.7BQ8_09.2 GBQ8_078details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_079details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_079details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_079details
Olmo 3 7B Instruct7.3BQ8_010.3 GBQ8_082details
Olmo 3 1025 7B7.3BQ8_010.3 GBQ8_082details
Olmo 3 7B Think7.3BQ8_010.3 GBQ8_082details
Mistral 7B v0.17.2BQ8_08.6 GBQ8_083details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ8_083details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ8_083details
Falcon 7B7.2BQ8_07.7 GBQ8_083details
Falcon 7B Instruct7.2BQ8_07.4 GBQ8_087details
Granite 4.0 H Tiny6.9B (1.5B act.)F1613.6 GBQ8_0218details
OLMoE 1B 7B 09246.9B (1.3B act.)Q8_08.4 GBQ8_0469details
DeepSeek Coder 7B Instruct v1.56.9BQ8_011.2 GBno87details
DeepSeek Coder 6.7B Instruct6.7BQ8_011.1 GBno89details
DeepSeek Coder 6.7B Base6.7BQ8_011.1 GBno89details
Llama 2 7B HF6.7BQ8_011.1 GBno89details
Yi 6B 200K6.1BF1612.3 GBQ8_053details
Yi 6B6.1BF1612.3 GBQ8_053details
Molmo2 4B4.9BF1610.0 GBF1672details
WeVisOne VL 4B4.8BF1610.0 GBF1672details
Mage VL Cowork4.7BF1610.0 GBF1672details
Qianfan OCR4.7BF1610.1 GBF1672details
Qwen3.5 4B4.7BF169.2 GBF1674details
SingVL S1 4B4.5BF169.0 GBF1676details
LightOnOCR 3 4B4.5BF169.0 GBF1676details
Intern Decision 4B4.5BF169.0 GBF1676details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1679details
Gemma 3 4B IT4.3BF168.4 GBF1682details
MedGemma 1.5 4B IT4.3BF168.4 GBF1682details
MedGemma 4B IT4.3BF168.4 GBF1682details
TranslateGemma 4B IT4.3BF169.2 GBF1674details
Gemma 3 4B PT4.3BF168.4 GBF1682details
Phi 3.5 Vision Instruct4.1BF1611.2 GBQ4_K_M77details
Qwen3 4B4.0BF169.3 GBF1680details
Qwen3 4B Instruct 25074.0BF169.3 GBF1679details
Granite Vision 4.1 4B4.0BF167.6 GBF1694details
Phi 4 Mini Instruct3.8BF168.9 GBF1683details
Phi 3 Mini 4K Instruct3.8BF1610.6 GBQ5_K_M84details
Phi 3.5 Mini Instruct3.8BF1610.6 GBQ5_K_M84details
Phi 3 Mini 128K Instruct3.8BF1610.6 GBQ5_K_M84details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF1685details
Granite 4.2 3B3.7BF168.0 GBF1687details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF1687details
Raon 1 Nano3.5BF167.0 GBF16102details
Granite 4.1 3B3.4BF167.6 GBF1694details
Granite 4.0 Micro3.4BF167.6 GBF1694details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16309details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16272details
Tiny Aya Global3.3BF167.5 GBF1695details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16317details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16318details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16318details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16362details
Llama 3.2 3B Instruct3.2BF167.5 GBF1699details
Llama 3.2 3B3.2BF167.5 GBF16100details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1699details
d1 3B3.1BF165.8 GBF16118details
LFM2.5 VL 3B3.1BF165.8 GBF16118details
Qwen2.5 3B Instruct3.1BF166.7 GBF16104details
SmolLM3 3B3.1BF166.9 GBF16104details
SmolLM3 3B Base3.1BF166.9 GBF16104details
Phi 22.8BF168.2 GBQ8_0115details
LFM2.5 2.6B2.7BF1610.8 GBF1659details
Granite 3.3 2B Instruct2.5BF165.8 GBF16126details
MiniCPM5 2B2.5BF165.7 GBF16127details
WeVisOne VL 2B2.4BF165.4 GBF16157details
Cosmos Reason2 2B2.4BF165.4 GBF16157details
Qwen3.5 2B2.3BF164.6 GBF16164details
SmolVLM Instruct2.2BF165.4 GBF16176details
Qwen3 1.7B2.0BF165.4 GBF16157details
d1 2B1.9BF164.0 GBF16185details
LFM3 VL 2B1.9BF164.0 GBF16185details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16180details
LFM3 1.7B1.7BF167.2 GBF1692details
LFM3 1.7B Base1.7BF164.0 GBF16185details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16187details
SmolLM2 1.7B1.7BF165.2 GBF16187details
SmolLM 1.7B1.7BF165.2 GBF16187details
Granite 4.0 1B1.6BF164.3 GBF16196details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16207details
OLMo 2 0425 1B1.5BF164.4 GBF16215details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16238details
Llama 3.2 1B1.2BF163.2 GBF16258details
Llama 3.2 1B Instruct1.2BF163.2 GBF16258details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16137details
LFM2 1.2B1.2BF163.0 GBF16273details
HunyuanOCR1.1BF163.5 GBF16286details
PaddleOCR VL 1.6959MF162.5 GBF16334details
Qwen3.5 0.8B873MF162.4 GBF16411details
LightOnOCR 3 0.8B853MF162.4 GBF16422details
FastVLM 0.5B759MF162.2 GBF16422details
Qwen3 0.6B752MF163.0 GBF16424details
GOT OCR2.0716MF162.8 GBF16447details
Nemotron 3.5 Parse631MF161.9 GBF16490details
GOT OCR 2.0 HF561MF162.5 GBF16571details
SmolVLM2 500M Video Instruct507MF161.6 GBF16780details
SmolVLM 500M Instruct507MF161.6 GBF16780details
Qwen2.5 0.5B Instruct494MF161.7 GBF16648details
LFM2.5 VL 450M449MF161.3 GBF16900details
SmolLM2 360M Instruct362MF161.5 GBF16882details
SmolLM2 360M362MF161.5 GBF16882details
LFM2.5 350M354MF161.9 GBF16450details

Bigger models with CPU offload

These popular models don't fit in the RX 9070 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 · ~7.3–14 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~14–28 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q2_K, 128 GB RAM · ~8.0–14 tok/s
DeepSeek V4 Flash291B MoEOffload Q3_K_M, 128 GB RAM · ~7.1–14 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 64 GB RAM · ~5.6–11 tok/s
MiniMax M2.5229B MoEOffload Q4_K_M, 128 GB RAM · ~5.6–11 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.2–8.3 tok/s

Comparing cards? The RX 9070 XT (launch MSRP $599) 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 RX 9070 XT has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
Trinity Mini15.8 GB0.6 GB
GLiDE 27B16.2 GB1.0 GB
MiniMax H3 Image PE16.8 GB1.6 GB
DiffusionGemma 26B A4B IT17.1 GB1.9 GB
Gemma 4 26B A4B IT17.3 GB2.1 GB
Gemma 3 27B IT17.3 GB2.1 GB
Arete 27B17.4 GB2.2 GB
Qwen3.8 27B17.6 GB2.4 GB
Muse Glimmer 30B17.9 GB2.7 GB
Qwen3.5 27B18.1 GB2.9 GB
Qwen3.6 27B18.1 GB2.9 GB
GLM 4.7 Flash18.3 GB3.1 GB
Hy Memory 30B A3B18.5 GB3.3 GB
Qwen3 Coder 30B A3B Instruct18.8 GB3.6 GB
North Mini Code 1.018.8 GB3.6 GB

Questions

What is the biggest model that runs on the RX 9070 XT?

Of the models we track, the largest that fits at 8K context is Maxwell V1 Flash (37B) at Q2_K, needing about 14.0 GB.

How much of the RX 9070 XT's memory can a model use?

About 15.2 GB: we assume 95% of the 16 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.