Best LLMs for 48 GB of VRAM
| Text models: 184 | Visual models: 12 | Sized for RTX PRO 5000 Blackwell 48 GB | Strongest first |
Text models for 48 GB
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.8 27B | 8-bit | 256K | 31 tok/s | Chat, Coding, Agents, Docs |
| Muse Glimmer 30B | 8-bit | 128K | 30 tok/s | Docs |
| Qwen3.5 122B A10B | 2-bit | 32K | 142 tok/s | Chat, Coding, Agents, Docs |
| Nemotron 3 Super 120B A12B | 2-bit | 184K | 98 tok/s | Chat, Coding, Agents, Docs |
| Nemotron 3.5 Bolt 74B A9B | 4-bit | 256K | 89 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.6 35B A3B | 8-bit | 256K | 171 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 31B | 8-bit | 256K | 27 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 26B A4B | 8-bit | 256K | 164 tok/s | Chat, Coding, Agents, Docs |
| GLM-4.7 Flash | 8-bit | 198K | 161 tok/s | Chat, Coding, Agents |
| Gemma 4 12B | 8-bit | 256K | 67 tok/s | Chat, Coding, Docs |
| Mistral Small 4 119B | 2-bit | 136K | 210 tok/s | Chat, Docs |
| gpt-oss 120B | 2-bit | 104K | 239 tok/s | Chat, Coding |
| Qwen3 Coder Next 80B | 3-bit | 256K | 221 tok/s | Coding, Agents |
| Nemotron 3.5 Lightning 30B A3B | 8-bit | 256K | 152 tok/s | Agents, Docs |
| Devstral Small 2 24B | FP8 | 128K | 36 tok/s | Coding |
| LFM2.5 24B A2B | 8-bit | 128K | 215 tok/s | Agents |
| Mistral Small 3.2 24B | 8-bit | 128K | 36 tok/s | Chat, Coding, Agents |
| Ministral 3 14B | 8-bit | 192K | 62 tok/s | Coding, Agents, Docs |
| gpt-oss 20B | MXFP4 | 128K | 240 tok/s | Agents |
| Ministral 3 8B | 8-bit | 256K | 95 tok/s | Docs |
| Llama 3.3 70B | 3-bit | 36K | 26 tok/s | Chat, Coding, Agents, Docs |
| Olmo 3.1 32B | 8-bit | 64K | 26 tok/s | Chat, Coding, Agents, Docs |
| Llama 3.1 8B | 8-bit | 128K | 106 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 14B | 8-bit | 40K | 59 tok/s | Not rated |
| Qwen3 VL 8B Instruct | 8-bit | 256K | 97 tok/s | Text and images |
| Qwen3 8B | 8-bit | 40K | 104 tok/s | Not rated |
| Qwen3 32B | 8-bit | 40K | 26 tok/s | Not rated |
| Qwen3.6 27B | 8-bit | 256K | 31 tok/s | Text and images |
| Qwen2.5 Coder 14B Instruct | 8-bit | 32K | 58 tok/s | Coding |
| Qwen3 Coder 30B A3B Instruct | 8-bit | 152K | 160 tok/s | Coding |
| Qwen3.5 27B | 8-bit | 256K | 31 tok/s | Text and images |
| Qwen3.5 35B A3B | 8-bit | 256K | 171 tok/s | Text and images |
| NVIDIA Nemotron 3 Nano 30B A3B BF16 | 8-bit | 256K | 152 tok/s | Not rated |
| GLM 4.5 Air | 2-bit | 32K | 115 tok/s | Not rated |
| Qwen3 30B A3B | 8-bit | 40K | 160 tok/s | Not rated |
| Qwen3 30B A3B Instruct 2507 | 8-bit | 152K | 160 tok/s | Not rated |
| Laguna XS.2 | 8-bit | 256K | 172 tok/s | Not rated |
| OTel 2.0 LLM 31B IT | 8-bit | 256K | 27 tok/s | Not rated |
| Qwen2.5 32B Instruct | 8-bit | 32K | 26 tok/s | Not rated |
| MiniCPM V 4.7 35B A3B | 8-bit | 256K | 171 tok/s | Text and images |
| Qwen2.5 14B Instruct | 8-bit | 32K | 58 tok/s | Not rated |
| Laguna XS 2.1 | 8-bit | 256K | 172 tok/s | Not rated |
| Diffusiongemma 26B A4B IT | 8-bit | 256K | 158 tok/s | Text and images |
| Qwen2.5 VL 32B Instruct | 8-bit | 44K | 26 tok/s | Text and images |
| Qwen2.5 Coder 32B Instruct | 8-bit | 32K | 26 tok/s | Coding |
| Laguna S 2.1 | 2-bit | 72K | 160 tok/s | Not rated |
| Qwen3 VL 30B A3B Instruct | 8-bit | 144K | 160 tok/s | Text and images |
| EXAONE 3.5 7.8B Instruct | 8-bit | 32K | 107 tok/s | Not rated |
| DeepSeek R1 0528 Qwen3 8B | 8-bit | 128K | 104 tok/s | Not rated |
| Gemma 3 12B IT | 8-bit | 128K | 71 tok/s | Text and images |
| Phi 3.5 Mini Instruct | 8-bit | 104K | 194 tok/s | Not rated |
| DeepSeek R1 Distill Qwen 32B | 8-bit | 48K | 26 tok/s | Not rated |
| Gemma 3 27B IT | 8-bit | 128K | 32 tok/s | Text and images |
| Ornith 1.5 35B A3B | 8-bit | 256K | 171 tok/s | Not rated |
| Ornith 1.5 122B | 2-bit | 32K | 123 tok/s | Not rated |
| GPT Neox 20B | 8-bit | 2K | 41 tok/s | Not rated |
| Apertus 8B Instruct 2509 | 8-bit | 64K | 106 tok/s | Not rated |
| Phi 4 | 8-bit | 16K | 58 tok/s | Not rated |
| Phi 3 Mini 128K Instruct | 8-bit | 104K | 194 tok/s | Not rated |
| Qwen3 VL 32B Instruct | 8-bit | 44K | 26 tok/s | Text and images |
Fits, context and speed on the RTX PRO 5000 Blackwell 48 GB. 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 184 are shown.
Image and video models
| Model | Best fit | Per output | Strongest at |
|---|---|---|---|
| MiniMax H3 | 16-bit | 4.0 min | Video |
| LTX-2.5 | 16-bit | 42 s | Video |
| FLUX.1 [dev] | 16-bit | 7.9 s | Images |
| Qwen-Image | 16-bit | 2.2 min | Images |
| FLUX.2 [dev] | 8-bit | 60 s | Images |
| Wan2.2 I2V A14B | 16-bit | 20 min | Video |
| Wan2.1 I2V 14B | 16-bit | 20 min | Video |
| Wan 2.1 T2V-14B | 16-bit | 25 min | Video |
| Krea 2 Turbo | 16-bit | 7.9 s | Images |
| Qwen Image 2512 | 16-bit | 2.2 min | Images |
| Wan 2.2 A14B | 16-bit | 20 min | Video |
| HunyuanVideo 1.5 | 16-bit | 24 min | Video |
Machines with 48 GB
Questions
What is the best LLM for 48 GB of VRAM?
Of the open models we track, the strongest that run well on 48 GB are Qwen3.8 27B, Muse Glimmer 30B, Qwen3.5 122B A10B. 184 text models fit in all.
Calculated from published specs, not benchmarked. Every machine and tier