What can the 2× RTX 5090 run?
| Text models: 130 | Visual models: 0 | Strongest first |
- Memory
- 64 GB
- Bandwidth
- 1,792 GB/s each
- Maker
- NVIDIA
Text models that run on it
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.8 27B | 8-bit | 256K | 41 tok/s | Chat, Coding, Agents, Docs |
| Muse Glimmer 30B | 8-bit | 128K | 39 tok/s | Docs |
| Mistral Medium 3.5 128B | 2-bit | 40K | 25 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.5 122B A10B | 3-bit | 80K | 145 tok/s | Chat, Coding, Agents, Docs |
| Nemotron 3 Super 120B A12B | 3-bit | 256K | 99 tok/s | Chat, Coding, Agents, Docs |
| Nemotron 3.5 Bolt 74B A9B | 6-bit | 248K | 89 tok/s | Chat, Coding, Agents, Docs |
| Qwen3.6 35B A3B | 8-bit | 256K | 200 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 31B | 8-bit | 256K | 35 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 26B A4B | 8-bit | 256K | 199 tok/s | Chat, Coding, Agents, Docs |
| GLM-4.7 Flash | 8-bit | 198K | 187 tok/s | Chat, Coding, Agents |
| Mistral Small 4 119B | 3-bit | 224K | 213 tok/s | Chat, Docs |
| gpt-oss 120B | 3-bit | 128K | 241 tok/s | Chat, Coding |
| Qwen3 Coder Next 80B | 5-bit | 256K | 215 tok/s | Coding, Agents |
| Nemotron 3.5 Lightning 30B A3B | 8-bit | 256K | 175 tok/s | Agents, Docs |
| Devstral Small 2 24B | FP8 | 224K | 47 tok/s | Coding |
| LFM2.5 24B A2B | 8-bit | 128K | 244 tok/s | Agents |
| Mistral Small 3.2 24B | 8-bit | 128K | 48 tok/s | Chat, Coding, Agents |
| Ministral 3 14B | 8-bit | 256K | 80 tok/s | Coding, Agents, Docs |
| Llama 3.3 70B | 6-bit | 16K | 21 tok/s | Chat, Coding, Agents, Docs |
| Olmo 3.1 32B | 8-bit | 64K | 35 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 14B | 8-bit | 40K | 77 tok/s | Not rated |
| Qwen3 32B | 8-bit | 40K | 34 tok/s | Not rated |
| Qwen3.6 27B | 8-bit | 256K | 41 tok/s | Text and images |
| Qwen2.5 Coder 14B Instruct | 8-bit | 32K | 75 tok/s | Coding |
| Qwen3 Coder 30B A3B Instruct | 8-bit | 256K | 184 tok/s | Coding |
| Qwen3.5 27B | 8-bit | 256K | 41 tok/s | Text and images |
| Qwen3.5 35B A3B | 8-bit | 256K | 200 tok/s | Text and images |
| NVIDIA Nemotron 3 Nano 30B A3B BF16 | 8-bit | 256K | 175 tok/s | Not rated |
| GLM 4.5 Air | 3-bit | 48K | 116 tok/s | Not rated |
| Qwen3 30B A3B | 8-bit | 40K | 184 tok/s | Not rated |
| Qwen3 30B A3B Instruct 2507 | 8-bit | 256K | 184 tok/s | Not rated |
| Laguna XS.2 | 8-bit | 256K | 202 tok/s | Not rated |
| OTel 2.0 LLM 31B IT | 8-bit | 256K | 35 tok/s | Not rated |
| Qwen2.5 32B Instruct | 8-bit | 32K | 34 tok/s | Not rated |
| MiniCPM V 4.7 35B A3B | 8-bit | 256K | 200 tok/s | Text and images |
| Qwen2.5 14B Instruct | 8-bit | 32K | 75 tok/s | Not rated |
| Laguna XS 2.1 | 8-bit | 256K | 202 tok/s | Not rated |
| Diffusiongemma 26B A4B IT | 8-bit | 256K | 193 tok/s | Text and images |
| Qwen2.5 VL 32B Instruct | 8-bit | 104K | 34 tok/s | Text and images |
| Qwen2.5 Coder 32B Instruct | 8-bit | 32K | 34 tok/s | Coding |
| Laguna S 2.1 | 3-bit | 120K | 162 tok/s | Not rated |
| Qwen3 VL 30B A3B Instruct | 8-bit | 256K | 184 tok/s | Text and images |
| Gemma 3 12B IT | 8-bit | 128K | 92 tok/s | Text and images |
| Phi 3.5 Mini Instruct | 8-bit | 128K | 243 tok/s | Not rated |
| DeepSeek R1 Distill Qwen 32B | 8-bit | 104K | 34 tok/s | Not rated |
| Gemma 3 27B IT | 8-bit | 128K | 42 tok/s | Text and images |
| Ornith 1.5 35B A3B | 8-bit | 256K | 200 tok/s | Not rated |
| Ornith 1.5 122B | 3-bit | 80K | 125 tok/s | Not rated |
| GPT Neox 20B | 8-bit | 2K | 54 tok/s | Not rated |
| Phi 4 | 8-bit | 16K | 76 tok/s | Not rated |
| Phi 3 Mini 128K Instruct | 8-bit | 128K | 243 tok/s | Not rated |
| Qwen3 VL 32B Instruct | 8-bit | 104K | 34 tok/s | Text and images |
| Pixtral 12B | 8-bit | 288K | 88 tok/s | Text and images |
| DeepSeek R1 Distill Qwen 14B | 8-bit | 128K | 75 tok/s | Not rated |
| Qwen2.5 14B Instruct 1M | 8-bit | 232K | 75 tok/s | Not rated |
| Qwen2.5 72B Instruct | 5-bit | 32K | 23 tok/s | Not rated |
| Qwen3 Next 80B A3B Instruct | 5-bit | 224K | 215 tok/s | Not rated |
| DeepSeek V2 Lite | 8-bit | 160K | 258 tok/s | Not rated |
| Llama 4 Scout 17B 16E Instruct | 3-bit | 48K | 101 tok/s | Text and images |
| Llama Guard 4 12B | 8-bit | 248K | 93 tok/s | Text and images |
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 130 are shown.
Only just too big
- Qwen3.8 Flash-Next
- Qwen3.5 397B A17B
- Ornith 1.5 397B
- Step 3.5 Flash
- Le Triomphant ECE TW3
- TW3 JRGL V2
- Command A Plus 05 2026 Bf16
- Qwen1.5 72B
- Step 3.7 Flash
These need up to a quarter more memory than the 2× RTX 5090 has. Spilling into system RAM or a lower precision can get them running, slowly.
Questions
What is the best LLM for the 2× RTX 5090?
Of the open models we track, the strongest that run well on the 2× RTX 5090 are Qwen3.8 27B, Muse Glimmer 30B, Mistral Medium 3.5 128B. 130 text models fit it in all.
Calculated from published specs, not benchmarked. All machines