What can the 2× RTX 4090 run?
| Text models: 188 | Visual models: 0 | Strongest first |
- Memory
- 48 GB
- Bandwidth
- 1,008 GB/s each
- Maker
- NVIDIA
Text models that run on it
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.8 27B | 8-bit | 256K | 24 tok/s | Chat, Coding, Agents, Docs |
| Muse Glimmer 30B | 8-bit | 128K | 23 tok/s | Docs |
| Nemotron 3 Super 120B A12B | 2-bit | 96K | 84 tok/s | Chat, Coding, Agents, Docs |
| Nemotron 3.5 Bolt 74B A9B | 4-bit | 216K | 75 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.6 35B A3B | 8-bit | 256K | 143 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 31B | 8-bit | 256K | 20 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 26B A4B | 8-bit | 256K | 133 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.5 9B | 8-bit | 256K | 71 tok/s | Chat, Coding, Agents, Docs |
| GLM-4.7 Flash | 8-bit | 198K | 136 tok/s | Chat, Coding, Agents |
| Gemma 4 12B | 8-bit | 256K | 51 tok/s | Chat, Coding, Docs |
| Mistral Small 4 119B | 2-bit | 104K | 178 tok/s | Chat, Docs |
| gpt-oss 120B | 2-bit | 88K | 205 tok/s | Chat, Coding |
| Qwen3 Coder Next 80B | 3-bit | 256K | 197 tok/s | Coding, Agents |
| Nemotron 3.5 Lightning 30B A3B | 8-bit | 256K | 130 tok/s | Agents, Docs |
| Devstral Small 2 24B | FP8 | 128K | 27 tok/s | Coding |
| LFM2.5 24B A2B | 8-bit | 128K | 185 tok/s | Agents |
| Mistral Small 3.2 24B | 8-bit | 128K | 28 tok/s | Chat, Coding, Agents |
| Ministral 3 14B | 8-bit | 192K | 47 tok/s | Coding, Agents, Docs |
| gpt-oss 20B | MXFP4 | 128K | 197 tok/s | Agents |
| Ministral 3 8B | 8-bit | 256K | 73 tok/s | Docs |
| Llama 3.3 70B | 3-bit | 36K | 20 tok/s | Chat, Coding, Agents, Docs |
| Olmo 3.1 32B | 8-bit | 64K | 20 tok/s | Chat, Coding, Agents, Docs |
| Llama 3.1 8B | 8-bit | 128K | 82 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 14B | 8-bit | 40K | 45 tok/s | Not rated |
| Qwen3 VL 8B Instruct | 8-bit | 248K | 75 tok/s | Text and images |
| Qwen3 8B | 8-bit | 40K | 80 tok/s | Not rated |
| Qwen3 32B | 8-bit | 40K | 20 tok/s | Not rated |
| Qwen3.6 27B | 8-bit | 256K | 24 tok/s | Text and images |
| Qwen2.5 Coder 14B Instruct | 8-bit | 32K | 44 tok/s | Coding |
| Qwen3 Coder 30B A3B Instruct | 8-bit | 144K | 135 tok/s | Coding |
| Qwen3.5 27B | 8-bit | 256K | 24 tok/s | Text and images |
| Qwen3.5 35B A3B | 8-bit | 256K | 143 tok/s | Text and images |
| NVIDIA Nemotron 3 Nano 30B A3B BF16 | 8-bit | 256K | 130 tok/s | Not rated |
| Mistral 7B Instruct V0.2 | 8-bit | 32K | 86 tok/s | Not rated |
| GLM 4.5 Air | 2-bit | 28K | 93 tok/s | Not rated |
| Qwen3 30B A3B | 8-bit | 40K | 135 tok/s | Not rated |
| Qwen3 30B A3B Instruct 2507 | 8-bit | 144K | 135 tok/s | Not rated |
| Laguna XS.2 | 8-bit | 256K | 144 tok/s | Not rated |
| OTel 2.0 LLM 31B IT | 8-bit | 256K | 20 tok/s | Not rated |
| Qwen2.5 32B Instruct | 8-bit | 32K | 20 tok/s | Not rated |
| MiniCPM V 4.7 35B A3B | 8-bit | 256K | 143 tok/s | Text and images |
| Qwen2.5 14B Instruct | 8-bit | 32K | 44 tok/s | Not rated |
| Laguna XS 2.1 | 8-bit | 256K | 144 tok/s | Not rated |
| Diffusiongemma 26B A4B IT | 8-bit | 256K | 128 tok/s | Text and images |
| Qwen2.5 VL 32B Instruct | 8-bit | 40K | 20 tok/s | Text and images |
| Qwen2.5 Coder 32B Instruct | 8-bit | 32K | 20 tok/s | Coding |
| Laguna S 2.1 | 2-bit | 64K | 136 tok/s | Not rated |
| Ornith 1.5 9B | 8-bit | 256K | 71 tok/s | Not rated |
| Qwen3 VL 30B A3B Instruct | 8-bit | 136K | 135 tok/s | Text and images |
| EXAONE 3.5 7.8B Instruct | 8-bit | 32K | 83 tok/s | Not rated |
| DeepSeek R1 0528 Qwen3 8B | 8-bit | 128K | 80 tok/s | Not rated |
| Gemma 3 12B IT | 8-bit | 128K | 55 tok/s | Text and images |
| Phi 3.5 Mini Instruct | 8-bit | 104K | 152 tok/s | Not rated |
| DeepSeek R1 Distill Qwen 32B | 8-bit | 44K | 20 tok/s | Not rated |
| Gemma 3 27B IT | 8-bit | 128K | 24 tok/s | Text and images |
| Ornith 1.5 35B A3B | 8-bit | 256K | 143 tok/s | Not rated |
| GPT Neox 20B | 8-bit | 2K | 31 tok/s | Not rated |
| Apertus 8B Instruct 2509 | 8-bit | 64K | 81 tok/s | Not rated |
| Phi 4 | 8-bit | 16K | 44 tok/s | Not rated |
| Phi 3 Mini 128K Instruct | 8-bit | 104K | 152 tok/s | Not 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 188 are shown.
Only just too big
- Mistral Medium 3.5 128B
- Qwen3.5 122B A10B
- Ornith 1.5 122B
- Le Triomphant ECE TW3
- TW3 JRGL V2
- Qwen1.5 72B
- WizardLM 2 8x22B
These need up to a quarter more memory than the 2× RTX 4090 has. Spilling into system RAM or a lower precision can get them running, slowly.
Questions
What is the best LLM for the 2× RTX 4090?
Of the open models we track, the strongest that run well on the 2× RTX 4090 are Qwen3.8 27B, Muse Glimmer 30B, Nemotron 3 Super 120B A12B. 188 text models fit it in all.
Calculated from published specs, not benchmarked. All machines