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

What AI models run on the RTX 5090?

The RTX 5090 has 32 GB of VRAM (1,792 GB/s). About 30.4 GB is usable for a model. 218 of 298 open models we track fit at 8K context.

218models fit (8K ctx)
209still fit at 32K ctx
80Blargest that fits
1792 GB/smemory bandwidth

This card has 32 GB of VRAM. See every model that fits in 32 GB VRAM. Other 32 GB hardware: Mac M4 32GB, Mac M1 Pro 32GB.

Compare for local LLMs: RTX 5090 vs RTX 5080RTX 5090 vs RTX 4090RTX 5090 vs RTX 3090RTX 5090 vs Mac M4 Max 128GBRTX 5090 vs DGX Spark all comparisons

Using a specific tool? Ollama models that run on the RTX 5090 · LM Studio · llama.cpp offload

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)Details
Hunyuan A13B Instruct80B (13B act.)Q2_K28.6 GBno383details
Kimi Dev 72B73BQ2_K28.6 GBno66details
Apertus v1.5 70B72BQ2_K28.6 GBno66details
Llama 3.3 70B Instruct71BQ2_K27.7 GBno68details
Llama 3.1 70B Instruct71BQ2_K27.7 GBno68details
Meta Llama 3 70B Instruct71BQ2_K27.7 GBno68details
DeepSeek R1 Distill Llama 70B71BQ2_K27.7 GBno68details
Llama 3.1 70B71BQ2_K27.7 GBno68details
Hermes 4 70B71BQ2_K27.7 GBno68details
Kimi Linear 48B A3B Instruct49B (3.0B act.)Q3_K_M22.1 GBQ3_K_M1294details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)Q3_K_M22.5 GBQ3_K_M288details
Falcon 40B42BQ4_K_M25.2 GBQ3_K_M70details
Flint40B (1.3B act.)Q4_K_M24.5 GBQ4_K_M2250details
Maxwell V1 Flash37B (2.3B act.)Q5_K_M25.6 GBQ5_K_M1108details
Hermes 4.3 36B36BQ5_K_M26.5 GBQ3_K_M70details
Seed OSS 36B Instruct36BQ5_K_M26.5 GBQ3_K_M70details
Qwen3.6 35B A3B36B (3.0B act.)Q5_K_M25.2 GBQ5_K_M829details
Ornith 1.5 35B A3B36B (3.0B act.)Q5_K_M24.8 GBQ5_K_M842details
Qwen3.5 35B A3B36B (3.0B act.)Q5_K_M25.2 GBQ5_K_M829details
Sage A1.536B (4.7B act.)Q5_K_M24.6 GBQ5_K_M536details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q5_K_M25.7 GBQ5_K_M783details
Yi 34B 200K34BQ5_K_M25.0 GBQ4_K_M74details
Yi 34B34BQ5_K_M25.0 GBQ4_K_M74details
Yi 34B Chat34BQ5_K_M25.0 GBQ4_K_M74details
Yi 1.5 34B Chat34BQ5_K_M25.0 GBQ4_K_M74details
Nous Hermes 2 Yi 34B34BQ5_K_M25.6 GBQ3_K_M72details
DeepSeek R1 Distill Qwen 32B33BQ5_K_M24.4 GBQ4_K_M77details
Qwen3 32B33BQ5_K_M24.3 GBQ4_K_M77details
GLM Z1 32B 041433BQ6_K26.1 GBQ5_K_M67details
GLM 4 32B 041433BQ6_K26.1 GBQ5_K_M67details
Olmo 3 32B Think32BQ6_K26.5 GBQ5_K_M68details
Granite 4.0 H Small32B (8.8B act.)Q6_K25.4 GBQ6_K248details
Gemma 4 31B IT31BQ5_K_M24.0 GBQ4_K_M79details
Gemma 4 31B31BQ6_K26.8 GBQ4_K_M70details
GLM 4.7 Flash31B (4.6B act.)Q6_K24.2 GBQ6_K489details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q6_K24.8 GBQ6_K726details
North Mini Code 1.030B (3.3B act.)Q6_K26.0 GBQ6_K626details
Hy Memory 30B A3B30B (3.0B act.)Q6_K24.3 GBQ6_K728details
Muse Glimmer 30B30BQ6_K23.7 GBQ6_K73details
Granite 4.2 30B29BQ6_K25.0 GBQ4_K_M75details
Raon 1 Mini Preview29B (4.3B act.)Q6_K25.2 GBQ6_K462details
Qwen3.8 27B28BQ6_K23.6 GBQ6_K75details
Qwen3.6 27B28BQ6_K23.4 GBQ6_K76details
Qwen3.5 27B28BQ6_K23.4 GBQ6_K76details
Gemma 3 27B IT27BQ6_K22.6 GBQ6_K81details
Arete 27B27BQ6_K23.4 GBQ6_K76details
MiniMax H3 Image PE27BQ6_K21.9 GBQ6_K81details
Trinity Mini26B (2.0B act.)Q8_026.9 GBQ8_0857details
GLiDE 27B26BQ8_027.2 GBQ6_K65details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q8_026.4 GBQ6_K430details
Gemma 4 26B A4B IT26B (4.0B act.)Q8_026.4 GBQ6_K430details
LFM2.5 24B A2B24B (2.0B act.)Q8_024.5 GBQ8_0843details
ERNIE 4.5 21B A3B PT22BQ8_022.6 GBQ8_077details
ERNIE 4.5 21B A3B Thinking22BQ8_022.6 GBQ8_077details
gpt-oss Safeguard 20B22B (4.8B act.)F1613.8 GBF16583details
gpt-oss 20B21B (4.2B act.)F1613.8 GBF16648details
Kimi VL A3B Thinking 250616B (3.0B act.)Q8_016.8 GBQ8_0577details
Kimi VL A3B Thinking16B (3.0B act.)Q8_017.2 GBQ8_0562details
Kimi VL A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0577details
DeepSeek MoE 16B Base16B (2.8B act.)Q8_018.6 GBQ8_0596details
LLaDA2.1 Mini16B (1.4B act.)Q8_017.1 GBQ8_01177details
Moonlight 16B A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0562details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q8_016.4 GBQ8_0633details
DeepSeek V2 Lite16B (2.7B act.)Q8_016.4 GBQ8_0634details
DeepSeek V2 Lite Chat16B (2.7B act.)Q8_016.4 GBQ8_0633details
Qwen2.5 Coder 14B Instruct15BQ8_016.8 GBQ8_0114details
DeepSeek R1 Distill Qwen 14B15BQ8_016.8 GBQ8_0114details
Qwen3 14B15BQ8_016.6 GBQ8_0114details
Phi 415BQ8_016.7 GBQ8_0115details
Gemma 3 12B IT12BF1623.6 GBF1676details
Llama Guard 4 12B12BF1624.6 GBQ8_075details
KORMo Web 11B11BF1622.7 GBF1680details
Llama 3.2 11B Vision Instruct11BF1620.1 GBF1692details
Llama 3.2 11B Vision11BF1621.7 GBF1684details
GLM 4.6V Flash10BF1618.5 GBF1695details
GLM 4.1V 9B Thinking10BF1618.5 GBF1695details
Step3 VL 10B10BF1617.1 GBF16109details
SenseNova Loop 9B Preview9.7BF1617.8 GBF16100details
Qwen3.5 9B9.7BF1618.3 GBF1697details
Ornith 1.5 9B9.7BF1617.8 GBF16100details
SenseNova Loop 9B9.4BF1618.7 GBF1695details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF16100details
GLM 4 9B9.4BF1618.5 GBF1695details
Apertus v1.5 8B8.9BF1618.2 GBF16101details
InternLM3 8B Instruct8.8BF1617.4 GBF16102details
Granite 4.2 8B8.8BF1618.2 GBF16102details
Granite 4.1 8B8.8BF1618.2 GBF16102details
Granite 4.1 8B Llmon Preview8.8BF1618.2 GBF16102details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF16109details
Molmo2 8B8.7BF1617.1 GBF16109details
LFM2.5 8B A1B8.5B (1.0B act.)F1616.5 GBF16895details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF16118details
UI TARS 1.5 7B8.3BF1615.3 GBF16118details
olmOCR 2 7B 10258.3BF1615.3 GBF16118details
olmOCR 7B 0225 Preview8.3BF1616.6 GBF16108details
UI TARS 7B DPO8.3BF1615.3 GBF16118details
Qwen3 8B8.2BF1617.1 GBF16109details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF16109details
Granite 3.3 8B Instruct8.2BF1617.0 GBF16110details
Granite 3.1 8B Instruct8.2BF1617.0 GBF16110details
Granite 3.2 8B Instruct8.2BF1617.0 GBF16110details
Apertus 8B Instruct 25098.1BF1616.7 GBF16111details
Llama 3.1 8B Instruct8.0BF1616.6 GBF16112details
Llama 3.1 8B8.0BF1616.6 GBF16112details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF16112details
Llama Guard 3 8B8.0BF1616.6 GBF16112details
DeepSeek R1 Distill Llama 8B8.0BF1616.6 GBF16112details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF16112details
Meta Llama Guard 2 8B8.0BF1616.6 GBF16112details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF16112details
Aya Expanse 8B8.0BF1616.9 GBF16112details
Ling 3.0 Tiny7.9B (1.4B act.)F1615.6 GBF16648details
MiMo 7B RL7.8BF1616.4 GBF16114details
InternLM2.5 7B Chat7.7BF1616.0 GBF16116details
InternLM2 Chat 7B7.7BF1616.0 GBF16116details
Qwen2.5 7B Instruct7.6BF1615.3 GBF16118details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF16118details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF16118details
Olmo 3 7B Instruct7.3BF1616.7 GBF16123details
Olmo 3 1025 7B7.3BF1616.7 GBF16123details
Olmo 3 7B Think7.3BF1616.7 GBF16123details
Mistral 7B v0.17.2BF1614.9 GBF16124details
Mistral 7B Instruct v0.27.2BF1614.9 GBF16124details
Mistral 7B Instruct v0.17.2BF1614.9 GBF16124details
Falcon 7B7.2BF1614.0 GBF16124details
Falcon 7B Instruct7.2BF1614.0 GBF16124details
Granite 4.0 H Tiny6.9B (1.5B act.)F1613.6 GBF16611details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF16699details
DeepSeek Coder 7B Instruct v1.56.9BF1617.2 GBQ8_0130details
DeepSeek Coder 6.7B Instruct6.7BF1617.0 GBQ8_0133details
DeepSeek Coder 6.7B Base6.7BF1617.0 GBQ8_0133details
Llama 2 7B HF6.7BF1617.0 GBQ8_0133details
Yi 6B 200K6.1BF1612.3 GBF16148details
Yi 6B6.1BF1612.3 GBF16148details
Molmo2 4B4.9BF1610.0 GBF16203details
WeVisOne VL 4B4.8BF1610.0 GBF16203details
Mage VL Cowork4.7BF1610.0 GBF16203details
Qianfan OCR4.7BF1610.1 GBF16203details
Qwen3.5 4B4.7BF169.2 GBF16207details
SingVL S1 4B4.5BF169.0 GBF16213details
LightOnOCR 3 4B4.5BF169.0 GBF16213details
Intern Decision 4B4.5BF169.0 GBF16213details
Qwen3 VL 4B Instruct4.4BF169.3 GBF16223details
Gemma 3 4B IT4.3BF168.4 GBF16231details
MedGemma 1.5 4B IT4.3BF168.4 GBF16231details
MedGemma 4B IT4.3BF168.4 GBF16231details
TranslateGemma 4B IT4.3BF169.2 GBF16208details
Gemma 3 4B PT4.3BF168.4 GBF16231details
Phi 3.5 Vision Instruct4.1BF1611.2 GBF16216details
Qwen3 4B4.0BF169.3 GBF16223details
Qwen3 4B Instruct 25074.0BF169.3 GBF16223details
Granite Vision 4.1 4B4.0BF167.6 GBF16263details
Phi 4 Mini Instruct3.8BF168.9 GBF16234details
Phi 3 Mini 4K Instruct3.8BF1610.6 GBF16234details
Phi 3.5 Mini Instruct3.8BF1610.6 GBF16234details
Phi 3 Mini 128K Instruct3.8BF1610.6 GBF16234details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF16239details
Granite 4.2 3B3.7BF168.0 GBF16245details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF16245details
Raon 1 Nano3.5BF167.0 GBF16286details
Granite 4.1 3B3.4BF167.6 GBF16263details
Granite 4.0 Micro3.4BF167.6 GBF16263details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16866details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16762details
Tiny Aya Global3.3BF167.5 GBF16267details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16886details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16892details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16892details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF161014details
Llama 3.2 3B Instruct3.2BF167.5 GBF16279details
Llama 3.2 3B3.2BF167.5 GBF16279details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF16279details
d1 3B3.1BF165.8 GBF16332details
LFM2.5 VL 3B3.1BF165.8 GBF16332details
Qwen2.5 3B Instruct3.1BF166.7 GBF16290details
SmolLM3 3B3.1BF166.9 GBF16291details
SmolLM3 3B Base3.1BF166.9 GBF16291details
Phi 22.8BF168.2 GBF16322details
LFM2.5 2.6B2.7BF1610.8 GBF16166details
Granite 3.3 2B Instruct2.5BF165.8 GBF16354details
MiniCPM5 2B2.5BF165.7 GBF16356details
WeVisOne VL 2B2.4BF165.4 GBF16440details
Cosmos Reason2 2B2.4BF165.4 GBF16440details
Qwen3.5 2B2.3BF164.6 GBF16460details
SmolVLM Instruct2.2BF165.4 GBF16494details
Qwen3 1.7B2.0BF165.4 GBF16440details
d1 2B1.9BF164.0 GBF16517details
LFM3 VL 2B1.9BF164.0 GBF16517details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16503details
LFM3 1.7B1.7BF167.2 GBF16258details
LFM3 1.7B Base1.7BF164.0 GBF16517details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16523details
SmolLM2 1.7B1.7BF165.2 GBF16524details
SmolLM 1.7B1.7BF165.2 GBF16523details
Granite 4.0 1B1.6BF164.3 GBF16548details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16580details
OLMo 2 0425 1B1.5BF164.4 GBF16603details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16665details
Llama 3.2 1B1.2BF163.2 GBF16723details
Llama 3.2 1B Instruct1.2BF163.2 GBF16723details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16382details
LFM2 1.2B1.2BF163.0 GBF16765details
HunyuanOCR1.1BF163.5 GBF16800details
PaddleOCR VL 1.6959MF162.5 GBF16935details
Qwen3.5 0.8B873MF162.4 GBF161150details
LightOnOCR 3 0.8B853MF162.4 GBF161181details
FastVLM 0.5B759MF162.2 GBF161181details
Qwen3 0.6B752MF163.0 GBF161187details
GOT OCR2.0716MF162.8 GBF161251details
Nemotron 3.5 Parse631MF161.9 GBF161371details
GOT OCR 2.0 HF561MF162.5 GBF161598details
SmolVLM2 500M Video Instruct507MF161.6 GBF162184details
SmolVLM 500M Instruct507MF161.6 GBF162184details
Qwen2.5 0.5B Instruct494MF161.7 GBF161814details
LFM2.5 VL 450M449MF161.3 GBF162519details
SmolLM2 360M Instruct362MF161.5 GBF162470details
SmolLM2 360M362MF161.5 GBF162470details
LFM2.5 350M354MF161.9 GBF161259details

Bigger models with CPU offload

These popular models don't fit in the RTX 5090'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 Q3_K_M, 128 GB RAM · ~6.3–13 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~22–44 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q3_K_M, 128 GB RAM · ~7.7–15 tok/s
DeepSeek V4 Flash291B MoEOffload Q3_K_M, 128 GB RAM · ~8.4–17 tok/s
MiniMax M2.5229B MoEOffload Q4_K_M, 128 GB RAM · ~6.6–14 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.9–10 tok/s

Comparing cards? The RTX 5090 (launch MSRP $1999) is in best GPUs for local LLMs under $2,000. See also Mac vs PC.

Questions

What is the biggest model that runs on the RTX 5090?

Of the models we track, the largest that fits at 8K context is Hunyuan A13B Instruct (80B) at Q2_K, needing about 28.6 GB.

How much of the RTX 5090's memory can a model use?

About 30.4 GB: we assume 95% of the 32 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.