Skip to content

What can the 4× RTX 3090 run?

What fits on the 4× RTX 3090
Text models: 119Visual models: 0Strongest first
Memory
96 GB
Bandwidth
936 GB/s each
Maker
NVIDIA

Open the calculator on this machine

Text models that run on it

Text models that run on the 4× RTX 3090, best first
ModelBest fitContextSpeedStrongest at
Qwen3.8 Flash-Next3-bit248K133 tok/sChat, Coding, Agents
MiniMax M31.6-bit48K92 tok/sDocs
Qwen3.8 27B8-bit256K22 tok/sChat, Coding, Agents, Docs
Qwen3.5 397B A17B1.6-bit256K108 tok/sAgents, Docs
MiniMax M2.72-bit32K106 tok/sCoding, Agents
Muse Glimmer 30B8-bit128K21 tok/sDocs
Mistral Medium 3.5 128B4-bit44K8.4 tok/sChat, Coding, Agents, Docs
Qwen3.5 122B A10B5-bit208K72 tok/sChat, Coding, Agents, Docs
Nemotron 3 Super 120B A12B5-bit256K54 tok/sChat, Coding, Agents, Docs
Nemotron 3.5 Bolt 74B A9B8-bit256K50 tok/sChat, Coding, Agents, Docs
Qwen3.6 35B A3B8-bit256K136 tok/sChat, Coding, Agents, Docs
Gemma 4 31B8-bit256K19 tok/sChat, Coding, Agents, Docs
Gemma 4 26B A4B8-bit256K126 tok/sChat, Coding, Agents, Docs
GLM-4.7 Flash8-bit198K130 tok/sChat, Coding, Agents
Mistral Small 4 119B5-bit256K109 tok/sChat, Docs
gpt-oss 120BMXFP4128K129 tok/sChat, Coding
Qwen3 Coder Next 80B8-bit256K127 tok/sCoding, Agents
Nemotron 3.5 Lightning 30B A3B8-bit256K124 tok/sAgents, Docs
Devstral Small 2 24BFP8384K25 tok/sCoding
LFM2.5 24B A2B8-bit128K177 tok/sAgents
Mistral Small 3.2 24B8-bit128K26 tok/sChat, Coding, Agents
Llama 3.3 70B8-bit56K8.7 tok/sChat, Coding, Agents, Docs
Olmo 3.1 32B8-bit64K19 tok/sChat, Coding, Agents, Docs
Qwen3 32B8-bit40K19 tok/sNot rated
Qwen3.6 27B8-bit256K22 tok/sText and images
Qwen3 Coder 30B A3B Instruct8-bit256K129 tok/sCoding
Qwen3.5 27B8-bit256K22 tok/sText and images
Qwen3.5 35B A3B8-bit256K136 tok/sText and images
NVIDIA Nemotron 3 Nano 30B A3B BF168-bit256K124 tok/sNot rated
GLM 4.5 Air6-bit20K47 tok/sNot rated
Qwen3 30B A3B8-bit40K129 tok/sNot rated
MiniMax M2.52-bit32K97 tok/sNot rated
Qwen3 30B A3B Instruct 25078-bit256K129 tok/sNot rated
Laguna XS.28-bit256K137 tok/sNot rated
OTel 2.0 LLM 31B IT8-bit256K19 tok/sNot rated
Laguna M.12-bit32K57 tok/sNot rated
Qwen2.5 32B Instruct8-bit32K19 tok/sNot rated
MiniCPM V 4.7 35B A3B8-bit256K136 tok/sText and images
Laguna XS 2.18-bit256K137 tok/sNot rated
Diffusiongemma 26B A4B IT8-bit256K121 tok/sText and images
Qwen2.5 VL 32B Instruct8-bit125K18 tok/sText and images
Qwen2.5 Coder 32B Instruct8-bit32K19 tok/sCoding
Laguna S 2.15-bit224K83 tok/sNot rated
Qwen3 VL 30B A3B Instruct8-bit256K129 tok/sText and images
DeepSeek R1 Distill Qwen 32B8-bit128K19 tok/sNot rated
Gemma 3 27B IT8-bit128K23 tok/sText and images
Qwen3 VL 235B A22B Instruct2-bit32K55 tok/sText and images
Qwen3 235B A22B2-bit32K55 tok/sNot rated
Ornith 1.5 35B A3B8-bit256K136 tok/sNot rated
Ornith 1.5 122B5-bit208K59 tok/sNot rated
GPT Neox 20B8-bit2K29 tok/sNot rated
Ornith 1.5 397B1.6-bit256K86 tok/sNot rated
Qwen3 VL 32B Instruct8-bit216K18 tok/sText and images
Step 3.5 Flash2-bit48K124 tok/sNot rated
Qwen2.5 72B Instruct8-bit32K8.5 tok/sNot rated
Qwen3 Next 80B A3B Instruct8-bit256K127 tok/sNot rated
Llama 4 Scout 17B 16E Instruct6-bit24K39 tok/sText and images
MiniMax M22-bit32K97 tok/sNot rated
Qwen2.5 VL 72B Instruct8-bit48K8.4 tok/sText and images
Qwen1.5 7B8-bit32K80 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 119 are shown.

Only just too big

These need up to a quarter more memory than the 4× RTX 3090 has. Spilling into system RAM or a lower precision can get them running, slowly.

Questions

What is the best LLM for the 4× RTX 3090?

Of the open models we track, the strongest that run well on the 4× RTX 3090 are Qwen3.8 Flash-Next, MiniMax M3, Qwen3.8 27B. 119 text models fit it in all.

Calculated from published specs, not benchmarked. All machines