Skip to content

Best LLMs for 24 GB of VRAM

What fits in 24 GB of VRAM
Text models: 258Visual models: 17Sized for RTX 3090Strongest first

Text models for 24 GB

Text models that fit 24 GB of VRAM, best first
ModelBest fitContextSpeedStrongest at
Qwen3.8 27B5-bit48K32 tok/sChat, Coding, Agents, Docs
Muse Glimmer 30B5-bit128K31 tok/sDocs
Qwen3.6 35B A3B4-bit72K190 tok/sChat, Coding, Agents, Docs
Gemma 4 31B4-bit80K33 tok/sChat, Coding, Agents, Docs
Gemma 4 26B A4B6-bit184K153 tok/sChat, Coding, Agents, Docs
Qwen3.5 9B8-bit256K66 tok/sChat, Coding, Agents, Docs
GLM-4.7 Flash5-bit24K165 tok/sChat, Coding, Agents
Gemma 4 12B8-bit256K48 tok/sChat, Coding, Docs
Nemotron 3.5 Lightning 30B A3B5-bit176K155 tok/sAgents, Docs
Qwen3.5 4B8-bit256K116 tok/sAgents, Docs
Devstral Small 2 24B6-bit20K33 tok/sCoding
LFM2.5 24B A2B6-bit128K204 tok/sAgents
Mistral Small 3.2 24B6-bit20K33 tok/sChat, Coding, Agents
Ministral 3 14B8-bit48K44 tok/sCoding, Agents, Docs
Gemma 4 E4B8-bit128K103 tok/sChat, Coding, Docs
gpt-oss 20BMXFP4128K187 tok/sAgents
Ministral 3 8B8-bit96K68 tok/sDocs
Gemma 4 E2B8-bit128K180 tok/sChat
Olmo 3.1 32B4-bit48K32 tok/sChat, Coding, Agents, Docs
Llama 3.1 8B8-bit112K76 tok/sChat, Coding, Agents, Docs
Qwen3 14B8-bit40K42 tok/sNot rated
Qwen3 VL 8B Instruct8-bit88K70 tok/sText and images
Qwen3 8B8-bit40K75 tok/sNot rated
Qwen2.5 7B Instruct8-bit32K81 tok/sNot rated
Qwen2.5 VL 7B Instruct8-bit125K75 tok/sText and images
Qwen3 VL 4B Instruct8-bit120K121 tok/sText and images
Qwen3 1.7B8-bit40K242 tok/sNot rated
Qwen3 4B8-bit40K132 tok/sNot rated
Qwen3 32B3-bit24K39 tok/sNot rated
Qwen3 4B Instruct 25078-bit128K132 tok/sNot rated
Qwen3.6 27B5-bit48K32 tok/sText and images
Qwen2.5 Coder 14B Instruct8-bit32K41 tok/sCoding
Qwen3 Coder 30B A3B Instruct5-bit16K163 tok/sCoding
Qwen3.5 27B5-bit48K32 tok/sText and images
Qwen3.5 35B A3B4-bit72K190 tok/sText and images
NVIDIA Nemotron 3 Nano 30B A3B BF165-bit176K155 tok/sNot rated
Qwen2.5 Coder 7B Instruct8-bit32K81 tok/sCoding
Mistral 7B Instruct V0.28-bit32K80 tok/sNot rated
Chandra OCR 28-bit256K104 tok/sReading documents
Qwen3 30B A3B5-bit16K163 tok/sNot rated
Qwen3 VL 2B Instruct8-bit176K233 tok/sText and images
Gemma 3 4B IT8-bit128K145 tok/sText and images
Qwen3 30B A3B Instruct 25075-bit16K163 tok/sNot rated
Llama 3.2 3B Instruct8-bit128K165 tok/sNot rated
Laguna XS.24-bit80K193 tok/sNot rated
OTel 2.0 LLM 31B IT4-bit80K33 tok/sNot rated
Qwen2.5 32B Instruct3-bit24K39 tok/sNot rated
MiniCPM V 4.7 35B A3B4-bit88K190 tok/sText and images
Qwen2.5 14B Instruct8-bit32K41 tok/sNot rated
Laguna XS 2.14-bit80K193 tok/sNot rated
Qwen3 1.7B Base8-bit32K275 tok/sNot rated
Diffusiongemma 26B A4B IT6-bit88K147 tok/sText and images
Qwen2.5 VL 32B Instruct3-bit24K38 tok/sText and images
Qwen2.5 Coder 32B Instruct3-bit24K39 tok/sCoding
Qwen2 VL 7B Instruct8-bit32K75 tok/sText and images
Ornith 1.5 9B8-bit256K66 tok/sNot rated
Granite 4.1 3B8-bit128K149 tok/sNot rated
Qwen3 VL 30B A3B Instruct4-bit44K177 tok/sText and images
Mistral 7B V0.18-bit32K80 tok/sNot rated
Meta Llama 3 8B Instruct Abliterated V38-bit8K76 tok/sNot rated

Fits, context and speed on the RTX 3090. 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 258 are shown.

Image and video models

Image and video models that fit 24 GB
ModelBest fitPer outputStrongest at
LTX-2.58-bit1.7 minVideo
Z-Image Turbo16-bit13 sImages
FLUX.1 [dev]16-bit26 sImages
Qwen-Image8-bit8.1 minImages
FLUX.2 [dev]5-bit4.1 minImages
Qwen-Image 2.116-bit3.5 minImages
FLUX.3 Klein16-bit19 sImages
Wan2.2 I2V A14B8-bit72 minVideo
FLUX.2 [klein] 9B16-bit6.8 sImages
Wan2.1 I2V 14B8-bit72 minVideo
Wan 2.1 T2V-14B8-bit90 minVideo
Krea 2 Turbo8-bit26 sImages
Stable Diffusion 3.5 Large16-bit1.8 minImages
Qwen Image 25128-bit8.1 minImages
Wan 2.2 TI2V-5B16-bit20 minVideo
Wan 2.2 A14B8-bit72 minVideo
HunyuanVideo 1.516-bit88 minVideo

Machines with 24 GB

Questions

What is the best LLM for 24 GB of VRAM?

Of the open models we track, the strongest that run well on 24 GB are Qwen3.8 27B, Muse Glimmer 30B, Qwen3.6 35B A3B. 258 text models fit in all.

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