Skip to content

Best LLMs for 32 GB of VRAM

What fits in 32 GB of VRAM
Text models: 233Visual models: 15Sized for RTX 5090Strongest first

Text models for 32 GB

Text models that fit 32 GB of VRAM, best first
ModelBest fitContextSpeedStrongest at
Qwen3.8 27B8-bit24K41 tok/sChat, Coding, Agents, Docs
Muse Glimmer 30B6-bit128K50 tok/sDocs
Qwen3.6 35B A3B6-bit96K227 tok/sChat, Coding, Agents, Docs
Gemma 4 31B6-bit112K45 tok/sChat, Coding, Agents, Docs
Gemma 4 26B A4B8-bit256K199 tok/sChat, Coding, Agents, Docs
Qwen3.5 9B8-bit256K120 tok/sChat, Coding, Agents, Docs
GLM-4.7 Flash6-bit104K209 tok/sChat, Coding, Agents
Gemma 4 12B8-bit256K86 tok/sChat, Coding, Docs
Nemotron 3.5 Lightning 30B A3B6-bit256K195 tok/sAgents, Docs
Devstral Small 2 24BFP832K47 tok/sCoding
LFM2.5 24B A2B8-bit128K244 tok/sAgents
Mistral Small 3.2 24B8-bit36K48 tok/sChat, Coding, Agents
Ministral 3 14B8-bit96K80 tok/sCoding, Agents, Docs
Gemma 4 E4B8-bit128K177 tok/sChat, Coding, Docs
gpt-oss 20BMXFP4128K287 tok/sAgents
Ministral 3 8B8-bit152K123 tok/sDocs
Olmo 3.1 32B6-bit64K44 tok/sChat, Coding, Agents, Docs
Llama 3.1 8B8-bit128K137 tok/sChat, Coding, Agents, Docs
Qwen3 14B8-bit40K77 tok/sNot rated
Qwen3 VL 8B Instruct8-bit144K125 tok/sText and images
Qwen3 8B8-bit40K134 tok/sNot rated
Qwen2.5 7B Instruct8-bit32K146 tok/sNot rated
Qwen2.5 VL 7B Instruct8-bit125K135 tok/sText and images
Qwen3 VL 4B Instruct8-bit176K207 tok/sText and images
Qwen3 4B8-bit40K223 tok/sNot rated
Qwen3 32B6-bit16K44 tok/sNot rated
Qwen3 4B Instruct 25078-bit176K223 tok/sNot rated
Qwen3.6 27B8-bit24K41 tok/sText and images
Qwen2.5 Coder 14B Instruct8-bit32K75 tok/sCoding
Qwen3 Coder 30B A3B Instruct6-bit64K207 tok/sCoding
Qwen3.5 27B8-bit24K41 tok/sText and images
Qwen3.5 35B A3B6-bit96K227 tok/sText and images
NVIDIA Nemotron 3 Nano 30B A3B BF166-bit256K195 tok/sNot rated
Qwen2.5 Coder 7B Instruct8-bit32K146 tok/sCoding
Mistral 7B Instruct V0.28-bit32K143 tok/sNot rated
Qwen3 30B A3B6-bit40K207 tok/sNot rated
Qwen3 30B A3B Instruct 25076-bit64K207 tok/sNot rated
Laguna XS.26-bit104K229 tok/sNot rated
OTel 2.0 LLM 31B IT6-bit112K45 tok/sNot rated
Qwen2.5 32B Instruct6-bit16K44 tok/sNot rated
MiniCPM V 4.7 35B A3B6-bit120K227 tok/sText and images
Qwen2.5 14B Instruct8-bit32K75 tok/sNot rated
Laguna XS 2.16-bit104K229 tok/sNot rated
Diffusiongemma 26B A4B IT8-bit184K193 tok/sText and images
Qwen2.5 VL 32B Instruct5-bit28K49 tok/sText and images
Qwen2.5 Coder 32B Instruct6-bit16K44 tok/sCoding
Qwen2 VL 7B Instruct8-bit32K135 tok/sText and images
Ornith 1.5 9B8-bit256K120 tok/sNot rated
Qwen3 VL 30B A3B Instruct6-bit60K207 tok/sText and images
Mistral 7B V0.18-bit32K143 tok/sNot rated
Meta Llama 3 8B Instruct Abliterated V38-bit8K137 tok/sNot rated
EXAONE 3.5 7.8B Instruct8-bit32K138 tok/sNot rated
Saiga Llama3 8B8-bit8K137 tok/sNot rated
DeepSeek R1 0528 Qwen3 8B8-bit128K134 tok/sNot rated
Gemma 3 12B IT8-bit128K92 tok/sText and images
Phi 3.5 Mini Instruct8-bit64K243 tok/sNot rated
DeepSeek R1 Distill Qwen 32B6-bit16K44 tok/sNot rated
Qwen3 4B Thinking 25078-bit176K223 tok/sNot rated
Gemma 3 27B IT8-bit20K42 tok/sText and images
Phi 4 Mini Instruct8-bit128K238 tok/sNot rated

Fits, context and speed on the RTX 5090. A slower card with the same memory holds the same models at lower speed. Ranked by our use-case scores, then by Hugging Face downloads. The top 60 of 233 are shown.

Image and video models

Image and video models that fit 32 GB
ModelBest fitPer outputStrongest at
MiniMax H38-bit4.9 minVideo
LTX-2.58-bit47 sVideo
FLUX.1 [dev]16-bit9.5 sImages
Qwen-Image8-bit2.7 minImages
FLUX.2 [dev]6-bit85 sImages
Qwen-Image 2.116-bit72 sImages
Wan2.2 I2V A14B8-bit24 minVideo
FLUX.2 [klein] 9B16-bit3.0 sImages
Wan2.1 I2V 14B8-bit24 minVideo
Wan 2.1 T2V-14B8-bit31 minVideo
Krea 2 Turbo16-bit9.4 sImages
Stable Diffusion 3.5 Large16-bit38 sImages
Qwen Image 25128-bit2.7 minImages
Wan 2.2 A14B8-bit24 minVideo
HunyuanVideo 1.516-bit30 minVideo

Machines with 32 GB

Questions

What is the best LLM for 32 GB of VRAM?

Of the open models we track, the strongest that run well on 32 GB are Qwen3.8 27B, Muse Glimmer 30B, Qwen3.6 35B A3B. 233 text models fit in all.

Calculated from published specs, not benchmarked. Every machine and tier