RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

What AI models run on the RX 7800 XT?

The RX 7800 XT has 16 GB of VRAM (624 GB/s). About 15.2 GB is usable for a model. 194 of 285 open models we track fit at 8K context.

194models fit (8K ctx)
180still fit at 32K ctx
40Blargest that fits
624 GB/smemory bandwidth

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)
flint40B (1.3B act.)Q2_K14.8 GBno1331details
Maxwell V1 Flash37B (2.3B act.)Q2_K14.0 GBQ2_K733details
Qwen3.6 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K555details
Ornith 1.5 35B A3B36B (3.0B act.)Q2_K13.2 GBQ2_K571details
Qwen3.5 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K555details
Axion 1.0 35B35B (3.9B act.)Q2_K13.3 GBQ2_K426details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q2_K13.1 GBQ2_K555details
Yi 1.5 34B Chat34BQ2_K14.3 GBno49details
Nous Hermes 2 Yi 34B34BQ2_K14.8 GBno47details
DeepSeek R1 Distill Qwen 32B33BQ2_K14.2 GBno51details
Qwen3 32B33BQ2_K14.1 GBno51details
GLM Z1 32B 041433BQ2_K12.6 GBQ2_K51details
GLM 4 32B 041433BQ2_K12.6 GBQ2_K51details
granite 4.0 h small32B (8.8B act.)Q2_K11.7 GBQ2_K194details
gemma 4 31B it31BQ2_K13.9 GBno53details
GLM 4.7 Flash31B (4.6B act.)Q2_K11.7 GBQ2_K373details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q2_K11.9 GBQ2_K564details
North Mini Code 1.030B (3.3B act.)Q2_K12.0 GBQ2_K503details
Hy Memory 30B A3B30B (3.0B act.)Q2_K11.9 GBQ2_K555details
Muse Glimmer 30B30BQ2_K11.4 GBQ2_K56details
granite 4.2 30b29BQ2_K12.7 GBno57details
Qwen3.8 27B28BQ2_K11.1 GBQ2_K60details
Qwen3.6 27B28BQ2_K11.1 GBQ2_K60details
Qwen3.5 27B28BQ2_K11.1 GBQ2_K60details
gemma 3 27b it27BQ2_K11.7 GBQ2_K59details
Arete 27B27BQ3_K_M13.6 GBQ2_K47details
MiniMax H3 Image PE27BQ2_K11.3 GBQ2_K58details
Trinity Mini26B (2.0B act.)Q3_K_M12.3 GBQ3_K_M685details
GLiDE 27B26BQ3_K_M13.2 GBQ2_K49details
diffusiongemma 26B A4B it26B (4.0B act.)Q3_K_M13.2 GBQ2_K319details
gemma 4 26B A4B it26B (4.0B act.)Q3_K_M13.3 GBQ2_K316details
LFM2.5 24B A2B24B (2.0B act.)Q3_K_M11.7 GBQ3_K_M638details
ERNIE 4.5 21B A3B PT22BQ4_K_M13.5 GBQ3_K_M47details
ERNIE 4.5 21B A3B Thinking22BQ4_K_M13.5 GBQ3_K_M47details
gpt oss safeguard 20b22B (4.8B act.)Q8_012.3 GBQ8_0231details
gpt oss 20b21B (4.2B act.)Q8_012.3 GBQ8_0257details
Groot 19B A3B 26102319B (3.0B act.)Q4_K_M12.7 GBQ4_K_M324details
Kimi VL A3B Thinking 250616B (3.0B act.)Q6_K13.5 GBQ5_K_M254details
Kimi VL A3B Instruct16B (3.0B act.)Q5_K_M12.2 GBQ5_K_M283details
Kimi VL A3B Thinking16B (3.0B act.)Q6_K13.5 GBQ5_K_M254details
deepseek moe 16b base16B (2.8B act.)Q4_K_M12.5 GBQ2_K333details
LLaDA2.1 mini16B (1.4B act.)Q6_K13.4 GBQ5_K_M531details
Moonlight 16B A3B Instruct16B (3.0B act.)Q5_K_M12.2 GBQ5_K_M276details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q5_K_M11.9 GBQ5_K_M311details
DeepSeek V2 Lite16B (2.7B act.)Q5_K_M11.3 GBQ5_K_M329details
DeepSeek V2 Lite Chat16B (2.7B act.)Q5_K_M11.9 GBQ5_K_M311details
Qwen2.5 Coder 14B Instruct15BQ6_K13.5 GBQ3_K_M52details
DeepSeek R1 Distill Qwen 14B15BQ6_K13.5 GBQ3_K_M51details
Qwen3 14B15BQ6_K13.2 GBQ3_K_M51details
phi 415BQ6_K13.4 GBQ3_K_M52details
gemma 3 12b it12BQ8_013.4 GBQ6_K50details
Llama Guard 4 12B12BQ6_K11.5 GBQ4_K_M63details
Llama 3.2 11B Vision Instruct11BQ8_011.6 GBQ5_K_M60details
Llama 3.2 11B Vision11BQ8_012.4 GBQ5_K_M55details
GLM 4.6V Flash10BQ8_010.3 GBQ8_062details
GLM 4.1V 9B Thinking10BQ8_010.3 GBQ8_062details
Step3 VL 10B10BQ8_09.9 GBQ8_072details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ8_065details
Qwen3.5 9B9.7BQ8_010.0 GBQ8_065details
Ornith 1.5 9B9.7BQ8_010.0 GBQ8_065details
Qwen3.5 9B Base9.7BQ8_010.0 GBQ8_065details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ8_065details
glm 4 9b9.4BQ6_K13.4 GBno76details
internlm3 8b instruct8.8BQ8_09.7 GBQ8_067details
granite 4.2 8b8.8BQ8_010.5 GBQ6_K67details
granite 4.1 8b8.8BQ8_010.5 GBQ6_K67details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ8_072details
Molmo2 8B8.7BQ8_09.9 GBQ8_072details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0586details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_077details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_077details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_077details
olmOCR 7B 0225 preview8.3BQ8_08.7 GBQ8_077details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_077details
Qwen3 8B8.2BQ8_09.9 GBQ8_072details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ8_072details
granite 3.3 8b instruct8.2BQ8_09.8 GBQ8_072details
granite 3.1 8b instruct8.2BQ8_09.8 GBQ8_072details
granite 3.2 8b instruct8.2BQ8_09.8 GBQ8_072details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ8_073details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_073details
Llama 3.1 8B8.0BQ8_09.6 GBQ8_073details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ8_073details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ8_073details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ8_073details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ8_073details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_073details
Ling 3.0 tiny7.9B (1.4B act.)Q8_08.7 GBQ8_0424details
MiMo 7B RL7.8BQ8_09.4 GBQ8_077details
internlm2 5 7b chat7.7BQ8_09.2 GBQ8_076details
internlm2 chat 7b7.7BQ8_09.2 GBQ8_076details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_077details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_077details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_077details
Falcon H1R 7B7.6BQ8_08.5 GBQ8_077details
Olmo 3 7B Instruct7.3BQ8_010.3 GBQ8_080details
Olmo 3 1025 7B7.3BQ8_010.3 GBQ8_080details
Olmo 3 7B Think7.3BQ8_010.3 GBQ8_080details
Mistral 7B v0.17.2BQ8_08.6 GBQ8_081details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ8_081details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ8_081details
falcon 7b7.2BQ8_012.1 GBno81details
falcon 7b instruct7.2BQ8_011.8 GBno85details
granite 4.0 h tiny6.9B (1.5B act.)F1613.6 GBQ8_0213details
OLMoE 1B 7B 09246.9B (1.3B act.)Q8_08.4 GBQ8_0458details
deepseek coder 7b instruct v1.56.9BQ8_011.2 GBno85details
deepseek coder 6.7b instruct6.7BQ8_011.1 GBno87details
deepseek coder 6.7b base6.7BQ8_011.1 GBno87details
Llama 2 7b hf6.7BQ8_011.1 GBno87details
Yi 6B 200K6.1BF1612.3 GBQ8_051details
Yi 6B6.1BF1612.3 GBQ8_051details
Molmo2 4B4.9BF1610.0 GBF1671details
WeVisOne VL 4B4.8BF1610.0 GBF1671details
Mage VL Cowork4.7BF1610.0 GBF1671details
Qianfan OCR4.7BF1610.1 GBF1671details
Qwen3.5 4B4.7BF169.0 GBF1674details
S5 nano4.7BF169.2 GBF1672details
LightOnOCR 3 4B4.5BF169.0 GBF1674details
SingVL S1 4B4.5BF169.0 GBF1674details
Intern Decision 4B4.5BF169.0 GBF1674details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1678details
gemma 3 4b it4.3BF168.4 GBF1680details
medgemma 1.5 4b it4.3BF168.4 GBF1680details
medgemma 4b it4.3BF168.4 GBF1680details
gemma 3 4b pt4.3BF169.2 GBF1673details
Phi 3.5 vision instruct4.1BF1611.2 GBQ4_K_M75details
Qwen3 4B4.0BF169.3 GBF1678details
Qwen3 4B Instruct 25074.0BF169.3 GBF1678details
granite vision 4.1 4b4.0BF167.6 GBF1692details
Phi 4 mini instruct3.8BF168.9 GBF1681details
Phi 3 mini 4k instruct3.8BF1610.6 GBQ5_K_M82details
Phi 3.5 mini instruct3.8BF1610.6 GBQ5_K_M82details
Phi 3 mini 128k instruct3.8BF1610.6 GBQ5_K_M82details
Qwen2.5 VL 3B Instruct3.8BF166.7 GBF16101details
granite 4.2 3b3.7BF168.0 GBF1685details
granite 4.1 3b llmon preview3.7BF168.0 GBF1685details
Raon 1 nano3.5BF167.0 GBF1699details
granite 4.1 3b3.4BF167.6 GBF1692details
granite 4.0 micro3.4BF167.6 GBF1692details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16301details
deepseek vl2 tiny3.4B (1.2B act.)F167.4 GBF16265details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16309details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16310details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16310details
granite 3.1 3b a800m instruct3.3B (883M act.)F167.1 GBF16353details
Llama 3.2 3B Instruct3.2BF1613.5 GBQ8_048details
Llama 3.2 3B3.2BF167.5 GBF1697details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1697details
d1 3B3.1BF1610.8 GBF1658details
LFM2.5 VL 3B3.1BF1610.8 GBF1658details
Qwen2.5 3B Instruct3.1BF166.7 GBF16101details
SmolLM3 3B3.1BF166.9 GBF16101details
SmolLM3 3B Base3.1BF166.9 GBF16101details
phi 22.8BF168.2 GBQ8_0112details
LFM2.5 2.6B2.7BF1610.8 GBF1658details
granite 3.3 2b instruct2.5BF165.8 GBF16123details
MiniCPM5 2B2.5BF165.7 GBF16124details
WeVisOne VL 2B2.4BF165.4 GBF16153details
Qwen3.5 2B2.3BF164.5 GBF16165details
SmolVLM Instruct2.2BF165.4 GBF16172details
Qwen3 1.7B2.0BF164.8 GBF16181details
Qwen3.8 27B DSpine b162.0BF164.8 GBF16154details
d1 2B1.9BF167.2 GBF1690details
LFM3 VL 2B1.9BF167.2 GBF1690details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16175details
LFM3 1.7B1.7BF167.2 GBF1690details
LFM3 1.7B Base1.7BF164.0 GBF16180details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16182details
SmolLM2 1.7B1.7BF165.2 GBF16182details
SmolLM 1.7B1.7BF165.2 GBF16182details
granite 4.0 1b1.6BF164.3 GBF16191details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16202details
OLMo 2 0425 1B1.5BF164.4 GBF16210details
deepseek coder 1.3b instruct1.3BF164.5 GBF16232details
Llama 3.2 1B1.2BF163.2 GBF16252details
Llama 3.2 1B Instruct1.2BF165.5 GBF16126details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16133details
HunyuanOCR1.1BF163.5 GBF16279details
PaddleOCR VL 1.6959MF162.5 GBF16325details
Qwen3.5 0.8B873MF162.4 GBF16411details
LightOnOCR 3 0.8B853MF162.4 GBF16411details
FastVLM 0.5B759MF162.2 GBF16411details
Qwen3 0.6B752MF162.7 GBF16521details
GOT OCR2 0716MF162.8 GBF16436details
Nemotron 3.5 Parse631MF161.9 GBF16477details
GOT OCR 2.0 hf561MF162.5 GBF16557details
SmolVLM2 500M Video Instruct507MF161.6 GBF16761details
SmolVLM 500M Instruct507MF161.6 GBF16761details
Qwen2.5 0.5B Instruct494MF161.7 GBF16632details
LFM2.5 VL 450M449MF161.9 GBF16439details
SmolLM2 360M Instruct362MF161.5 GBF16860details
SmolLM2 360M362MF161.5 GBF16860details
LFM2.5 350M354MF161.9 GBF16439details
LFM2 350M354MF161.4 GBF16880details

Just too big at 4-bit

These need up to 35% more memory than the RX 7800 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
Qwen3.8 27B16.8 GB1.6 GB
Qwen3.5 27B17.0 GB1.8 GB
Qwen3.6 27B17.0 GB1.8 GB
diffusiongemma 26B A4B it17.1 GB1.9 GB
gemma 4 26B A4B it17.2 GB2.0 GB
gemma 3 27b it17.3 GB2.1 GB
Arete 27B17.4 GB2.2 GB
Muse Glimmer 30B17.9 GB2.7 GB
GLM 4.7 Flash18.2 GB3.0 GB
Hy Memory 30B A3B18.5 GB3.3 GB
Qwen3 Coder 30B A3B Instruct18.7 GB3.5 GB
granite 4.0 h small18.9 GB3.7 GB

Questions

What is the biggest model that runs on the RX 7800 XT?

Of the models we track, the largest that fits at 8K context is flint (40B) at Q2_K, needing about 14.8 GB.

How much of the RX 7800 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.