RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

What AI models run on the DGX Spark?

The DGX Spark has 128 GB of unified memory (273 GB/s). About 109 GB is usable for a model. 235 of 285 open models we track fit at 8K context.

235models fit (8K ctx)
229still fit at 32K ctx
310Blargest that fits
273 GB/smemory bandwidth

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)
MiMo V2 Flash310B (16B act.)Q2_K106 GBno46details
DeepSeek V4 Flash Vision Exp305B (15B act.)Q2_K108 GBno48details
DeepSeek V4 Flash 0731304B (27B act.)Q2_K108 GBno27details
DeepSeek V4 Flash DSpark304B (27B act.)Q2_K108 GBno27details
Hy3299B (21B act.)Q2_K102 GBno36details
Hy3 preview299B (21B act.)Q2_K107 GBno35details
DeepSeek V4 Flash291B (14B act.)Q2_K103 GBQ2_K52details
ERNIE 5.5 Flash Base286B (5.7B act.)Q2_K101 GBQ2_K—details
Intern S1241B (28B act.)Q2_K86.2 GBQ2_K26details
DeepSeek V2 Chat236B (21B act.)Q2_K83.5 GBQ2_K34details
MiniMax M2.5229B (11B act.)Q2_K80.3 GBQ2_K68details
MiniMax M2.7229B (11B act.)Q3_K_M97.0 GBQ2_K56details
MiniMax M2229B (11B act.)Q2_K80.3 GBQ2_K68details
MiniMax M2.1229B (11B act.)Q2_K80.3 GBQ2_K68details
MiniMax M3.1 Flash220B (4.4B act.)Q2_K78.7 GBQ2_K—details
Step 3.7 Flash201BQ3_K_M91.6 GBQ2_K2.9details
Step 3.5 Flash199BQ3_K_M88.2 GBQ3_K_M3.0details
Qwen3.8 Flash Next180B (62B act.)Q3_K_M83.0 GBQ3_K_M9.1details
Ling 3.0 flash Sante127B (5.6B act.)Q5_K_M92.0 GBQ5_K_M64details
Ling 3.0 flash127B (5.6B act.)Q5_K_M85.5 GBQ5_K_M68details
Ornith 1.5 122B125B (13B act.)Q6_K96.6 GBQ6_K26details
gpt oss 120b117B (5.8B act.)F1661.9 GBF1685details
gpt oss safeguard 120b117B (5.8B act.)Q8_060.1 GBQ8_087details
GLM 4.5 Air110B (15B act.)Q6_K94.3 GBQ5_K_M20details
Llama 4 Scout 17B 16E Instruct109B (18B act.)Q6_K84.6 GBQ6_K19details
Llama 4 Scout 17B 16E109B (18B act.)Q6_K85.3 GBQ6_K18details
GLM 4.5V108B (14B act.)Q6_K84.4 GBQ6_K23details
GLM 4.6V108B (14B act.)Q6_K91.4 GBQ6_K21details
Hunyuan A13B Instruct80B (13B act.)Q8_081.2 GBQ8_020details
Kimi Dev 72B73BQ8_075.1 GBQ8_03.5details
Llama 3.3 70B Instruct71BQ8_073.0 GBQ8_03.6details
DeepSeek R1 Distill Llama 70B71BQ8_073.0 GBQ8_03.6details
Llama 3.1 70B71BQ8_073.0 GBQ8_03.6details
Kimi Linear 48B A3B Instruct49B (3.0B act.)F1692.5 GBF1645details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)F1688.5 GBF1611details
falcon 40b42BF1679.4 GBF163.3details
flint40B (1.3B act.)F1675.1 GBF16109details
Maxwell V1 Flash37B (2.3B act.)F1669.7 GBF1660details
Hermes 4.3 36B36BF1670.0 GBF163.8details
Seed OSS 36B Instruct36BF1670.0 GBF163.8details
Qwen3.6 35B A3B36B (3.0B act.)F1667.2 GBF1646details
Ornith 1.5 35B A3B36B (3.0B act.)F1667.2 GBF1646details
Qwen3.5 35B A3B36B (3.0B act.)F1665.6 GBF1647details
Axion 1.0 35B35B (3.9B act.)F1665.6 GBF1635details
Qwen AgentWorld 35B A3B35B (3.0B act.)F1665.6 GBF1645details
Yi 34B 200K34BF1666.5 GBF164.0details
Yi 34B34BF1666.5 GBF164.0details
Yi 34B Chat34BF1666.5 GBF164.0details
Yi 1.5 34B Chat34BF1666.5 GBF164.0details
Nous Hermes 2 Yi 34B34BF1666.5 GBF164.0details
DeepSeek R1 Distill Qwen 32B33BF1663.7 GBF164.2details
Qwen3 32B33BF1663.7 GBF164.2details
GLM Z1 32B 041433BF1661.8 GBF164.2details
GLM 4 32B 041433BF1661.8 GBF164.2details
granite 4.0 h small32B (8.8B act.)F1660.7 GBF1615details
gemma 4 31B it31BF1660.1 GBF164.4details
GLM 4.7 Flash31B (4.6B act.)F1656.9 GBF1631details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)F1658.3 GBF1645details
North Mini Code 1.030B (3.3B act.)F1658.2 GBF1641details
Hy Memory 30B A3B30B (3.0B act.)F1657.4 GBF1646details
Muse Glimmer 30B30BF1652.9 GBF164.9details
granite 4.2 30b29BF1657.1 GBF164.7details
Qwen3.8 27B28BF1652.3 GBF165.0details
Qwen3.6 27B28BF1651.5 GBF165.1details
Qwen3.5 27B28BF1651.5 GBF165.1details
gemma 3 27b it27BF1652.2 GBF165.1details
Arete 27B27BF1651.5 GBF165.1details
MiniMax H3 Image PE27BF1652.3 GBF165.0details
Trinity Mini26B (2.0B act.)F1649.7 GBF1669details
GLiDE 27B26BF1650.0 GBF165.2details
diffusiongemma 26B A4B it26B (4.0B act.)F1648.5 GBF1635details
gemma 4 26B A4B it26B (4.0B act.)F1648.4 GBF1635details
LFM2.5 24B A2B24B (2.0B act.)F1690.1 GBF1634details
ERNIE 4.5 21B A3B PT22BF1641.7 GBF166.3details
ERNIE 4.5 21B A3B Thinking22BF1641.7 GBF166.3details
gpt oss safeguard 20b22B (4.8B act.)F1613.8 GBF1689details
gpt oss 20b21B (4.2B act.)F1613.8 GBF1699details
Groot 19B A3B 26102319B (3.0B act.)F1636.4 GBF1647details
Kimi VL A3B Thinking 250616B (3.0B act.)F1660.4 GBF1623details
Kimi VL A3B Instruct16B (3.0B act.)F1631.5 GBF1646details
Kimi VL A3B Thinking16B (3.0B act.)F1631.5 GBF1646details
deepseek moe 16b base16B (2.8B act.)F1632.8 GBF1648details
LLaDA2.1 mini16B (1.4B act.)F1631.3 GBF1695details
Moonlight 16B A3B Instruct16B (3.0B act.)F1630.7 GBF1646details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)F1630.1 GBF1651details
DeepSeek V2 Lite16B (2.7B act.)F1630.1 GBF1651details
DeepSeek V2 Lite Chat16B (2.7B act.)F1630.1 GBF1651details
Qwen2.5 Coder 14B Instruct15BF1629.7 GBF169.2details
DeepSeek R1 Distill Qwen 14B15BF1629.7 GBF169.2details
Qwen3 14B15BF1629.4 GBF169.2details
phi 415BF1629.5 GBF169.3details
gemma 3 12b it12BF1623.6 GBF1612details
Llama Guard 4 12B12BF1624.6 GBF1611details
Llama 3.2 11B Vision Instruct11BF1620.1 GBF1614details
Llama 3.2 11B Vision11BF1621.7 GBF1613details
GLM 4.6V Flash10BF1618.5 GBF1615details
GLM 4.1V 9B Thinking10BF1618.5 GBF1615details
Step3 VL 10B10BF1632.3 GBF168.3details
SenseNova Loop 9B Preview9.7BF1617.8 GBF1615details
Qwen3.5 9B9.7BF1617.8 GBF1615details
Ornith 1.5 9B9.7BF1617.8 GBF1615details
Qwen3.5 9B Base9.7BF1617.8 GBF1615details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF1615details
glm 4 9b9.4BF1623.2 GBF1615details
internlm3 8b instruct8.8BF1617.4 GBF1616details
granite 4.2 8b8.8BF1618.2 GBF1616details
granite 4.1 8b8.8BF1618.2 GBF1616details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF1617details
Molmo2 8B8.7BF1617.1 GBF1617details
LFM2.5 8B A1B8.5B (1.0B act.)F1632.3 GBF1668details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF1618details
UI TARS 1.5 7B8.3BF1615.3 GBF1618details
olmOCR 2 7B 10258.3BF1615.3 GBF1618details
olmOCR 7B 0225 preview8.3BF1616.6 GBF1616details
UI TARS 7B DPO8.3BF1615.3 GBF1618details
Qwen3 8B8.2BF1617.1 GBF1617details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF1617details
granite 3.3 8b instruct8.2BF1617.0 GBF1617details
granite 3.1 8b instruct8.2BF1617.0 GBF1617details
granite 3.2 8b instruct8.2BF1617.0 GBF1617details
Apertus 8B Instruct 25098.1BF1616.7 GBF1617details
Llama 3.1 8B Instruct8.0BF1616.6 GBF1617details
Llama 3.1 8B8.0BF1616.6 GBF1617details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF1617details
DeepSeek R1 Distill Llama 8B8.0BF1631.6 GBF168.5details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF1617details
Meta Llama Guard 2 8B8.0BF1616.6 GBF1617details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF1617details
Ling 3.0 tiny7.9B (1.4B act.)F1615.6 GBF1699details
MiMo 7B RL7.8BF1616.4 GBF1617details
internlm2 5 7b chat7.7BF1616.0 GBF1618details
internlm2 chat 7b7.7BF1616.0 GBF1618details
Qwen2.5 7B Instruct7.6BF1615.3 GBF1618details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF1618details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF1618details
Falcon H1R 7B7.6BF1615.1 GBF1618details
Olmo 3 7B Instruct7.3BF1616.7 GBF1619details
Olmo 3 1025 7B7.3BF1616.7 GBF1619details
Olmo 3 7B Think7.3BF1616.7 GBF1619details
Mistral 7B v0.17.2BF1614.9 GBF1619details
Mistral 7B Instruct v0.27.2BF1614.9 GBF1619details
Mistral 7B Instruct v0.17.2BF1614.9 GBF1619details
falcon 7b7.2BF1618.4 GBF1619details
falcon 7b instruct7.2BF1618.4 GBF1619details
granite 4.0 h tiny6.9B (1.5B act.)F1613.6 GBF1693details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF16106details
deepseek coder 7b instruct v1.56.9BF1617.2 GBF1620details
deepseek coder 6.7b instruct6.7BF1617.0 GBF1620details
deepseek coder 6.7b base6.7BF1617.0 GBF1620details
Llama 2 7b hf6.7BF1617.0 GBF1620details
Yi 6B 200K6.1BF1612.3 GBF1623details
Yi 6B6.1BF1612.3 GBF1623details
Molmo2 4B4.9BF1610.0 GBF1631details
WeVisOne VL 4B4.8BF1610.0 GBF1631details
Mage VL Cowork4.7BF1610.0 GBF1631details
Qianfan OCR4.7BF1610.1 GBF1631details
Qwen3.5 4B4.7BF169.0 GBF1632details
S5 nano4.7BF169.2 GBF1632details
LightOnOCR 3 4B4.5BF169.0 GBF1632details
SingVL S1 4B4.5BF169.0 GBF1632details
Intern Decision 4B4.5BF169.0 GBF1632details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1634details
gemma 3 4b it4.3BF168.4 GBF1635details
medgemma 1.5 4b it4.3BF168.4 GBF1635details
medgemma 4b it4.3BF168.4 GBF1635details
gemma 3 4b pt4.3BF169.2 GBF1632details
Phi 3.5 vision instruct4.1BF1611.2 GBF1633details
Qwen3 4B4.0BF169.3 GBF1634details
Qwen3 4B Instruct 25074.0BF169.3 GBF1634details
granite vision 4.1 4b4.0BF167.6 GBF1640details
Phi 4 mini instruct3.8BF168.9 GBF1636details
Phi 3 mini 4k instruct3.8BF1610.6 GBF1636details
Phi 3.5 mini instruct3.8BF1610.6 GBF1636details
Phi 3 mini 128k instruct3.8BF1610.6 GBF1636details
Qwen2.5 VL 3B Instruct3.8BF166.7 GBF1644details
granite 4.2 3b3.7BF168.0 GBF1637details
granite 4.1 3b llmon preview3.7BF168.0 GBF1637details
Raon 1 nano3.5BF167.0 GBF1644details
granite 4.1 3b3.4BF167.6 GBF1640details
granite 4.0 micro3.4BF167.6 GBF1640details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16132details
deepseek vl2 tiny3.4B (1.2B act.)F167.4 GBF16116details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16135details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16136details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16136details
granite 3.1 3b a800m instruct3.3B (883M act.)F167.1 GBF16154details
Llama 3.2 3B Instruct3.2BF1613.5 GBF1621details
Llama 3.2 3B3.2BF167.5 GBF1642details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1642details
d1 3B3.1BF1610.8 GBF1625details
LFM2.5 VL 3B3.1BF1610.8 GBF1625details
Qwen2.5 3B Instruct3.1BF166.7 GBF1644details
SmolLM3 3B3.1BF166.9 GBF1644details
SmolLM3 3B Base3.1BF166.9 GBF1644details
phi 22.8BF168.2 GBF1649details
LFM2.5 2.6B2.7BF1610.8 GBF1625details
granite 3.3 2b instruct2.5BF165.8 GBF1654details
MiniCPM5 2B2.5BF165.7 GBF1654details
WeVisOne VL 2B2.4BF165.4 GBF1667details
Qwen3.5 2B2.3BF164.5 GBF1672details
SmolVLM Instruct2.2BF165.4 GBF1675details
Qwen3 1.7B2.0BF164.8 GBF1679details
Qwen3.8 27B DSpine b162.0BF164.8 GBF1667details
d1 2B1.9BF167.2 GBF1639details
LFM3 VL 2B1.9BF167.2 GBF1639details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF1677details
LFM3 1.7B1.7BF167.2 GBF1639details
LFM3 1.7B Base1.7BF164.0 GBF1679details
SmolLM2 1.7B Instruct1.7BF165.2 GBF1680details
SmolLM2 1.7B1.7BF165.2 GBF1680details
SmolLM 1.7B1.7BF165.2 GBF1680details
granite 4.0 1b1.6BF164.3 GBF1684details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF1688details
OLMo 2 0425 1B1.5BF164.4 GBF1692details
deepseek coder 1.3b instruct1.3BF164.5 GBF16101details
Llama 3.2 1B1.2BF163.2 GBF16110details
Llama 3.2 1B Instruct1.2BF165.5 GBF1655details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1658details
HunyuanOCR1.1BF163.5 GBF16122details
PaddleOCR VL 1.6959MF162.5 GBF16142details
Qwen3.5 0.8B873MF162.4 GBF16180details
LightOnOCR 3 0.8B853MF162.4 GBF16180details
FastVLM 0.5B759MF162.2 GBF16180details
Qwen3 0.6B752MF162.7 GBF16228details
GOT OCR2 0716MF162.8 GBF16191details
Nemotron 3.5 Parse631MF161.9 GBF16209details
GOT OCR 2.0 hf561MF162.5 GBF16244details
SmolVLM2 500M Video Instruct507MF161.6 GBF16333details
SmolVLM 500M Instruct507MF161.6 GBF16333details
Qwen2.5 0.5B Instruct494MF161.7 GBF16276details
LFM2.5 VL 450M449MF161.9 GBF16192details
SmolLM2 360M Instruct362MF161.5 GBF16376details
SmolLM2 360M362MF161.5 GBF16376details
LFM2.5 350M354MF161.9 GBF16192details
LFM2 350M354MF161.4 GBF16385details

Just too big at 4-bit

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

ModelQ4_K_M needsShort by
Step 3.5 Flash115.9 GB7.1 GB
Step 3.7 Flash118.8 GB10.0 GB
MiniMax M3.1 Flash127.2 GB18.4 GB
MiniMax M2131.6 GB22.8 GB
MiniMax M2.5131.6 GB22.8 GB
MiniMax M2.1131.6 GB22.8 GB
MiniMax M2.7133.3 GB24.5 GB
DeepSeek V2 Chat135.3 GB26.5 GB
Intern S1139.2 GB30.4 GB

Questions

What is the biggest model that runs on the DGX Spark?

Of the models we track, the largest that fits at 8K context is MiMo V2 Flash (310B) at Q2_K, needing about 106 GB.

How much of the DGX Spark's memory can a model use?

About 109 GB of 128 GB (we assume ~85% of shared memory can be given to the GPU).

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.