What can the RTX 3070 run?
| Text models: 234 | Visual models: 6 | Strongest first |
- Memory
- 8 GB
- Bandwidth
- 448 GB/s
- Maker
- NVIDIA
Open the calculator on this machineEvery model for 8 GB of VRAM
Text models that run on it
| Model | Best fit | Context | Speed | Strongest at |
|---|---|---|---|---|
| Qwen3.5 9B | 4-bit | 32K | 56 tok/s | Chat, Coding, Agents, Docs |
| Gemma 4 12B | 3-bit | 96K | 50 tok/s | Chat, Coding, Docs |
| Qwen3.5 4B | 8-bit | 60K | 59 tok/s | Agents, Docs |
| Gemma 4 E4B | 6-bit | 28K | 67 tok/s | Chat, Coding, Docs |
| Ministral 3 8B | 3-bit | 20K | 71 tok/s | Docs |
| Gemma 4 E2B | 8-bit | 128K | 96 tok/s | Chat |
| Nemotron 3 Nano 4B | 8-bit | 176K | 74 tok/s | Docs |
| Llama 3.1 8B | 4-bit | 16K | 65 tok/s | Chat, Coding, Agents, Docs |
| Qwen3 0.6B | 8-bit | 40K | 295 tok/s | Not rated |
| Qwen3 VL 8B Instruct | 3-bit | 16K | 73 tok/s | Text and images |
| Qwen3 8B | 3-bit | 20K | 78 tok/s | Not rated |
| Qwen2.5 7B Instruct | 6-bit | 16K | 52 tok/s | Not rated |
| Qwen2.5 VL 7B Instruct | 5-bit | 20K | 54 tok/s | Text and images |
| Unlimited OCR | 8-bit | 32K | 218 tok/s | Reading documents |
| Qwen2.5 1.5B Instruct | 8-bit | 32K | 165 tok/s | Not rated |
| Qwen3.5 2B | 8-bit | 256K | 118 tok/s | Text and images |
| Qwen3 VL 4B Instruct | 8-bit | 16K | 62 tok/s | Text and images |
| Qwen3 1.7B | 8-bit | 40K | 129 tok/s | Not rated |
| Qwen3.5 0.8B | 8-bit | 256K | 267 tok/s | Text and images |
| Qwen3 4B | 8-bit | 16K | 68 tok/s | Not rated |
| Qwen3 4B Instruct 2507 | 8-bit | 16K | 68 tok/s | Not rated |
| Qwen2.5 3B Instruct | 8-bit | 32K | 87 tok/s | Not rated |
| DeepSeek OCR | 8-bit | 8K | 218 tok/s | Reading documents |
| Qwen2.5 VL 3B Instruct | 8-bit | 80K | 72 tok/s | Text and images |
| Qwen2.5 Coder 7B Instruct | 6-bit | 16K | 52 tok/s | Coding |
| Mistral 7B Instruct V0.2 | 5-bit | 16K | 59 tok/s | Not rated |
| Chandra OCR 2 | 8-bit | 40K | 52 tok/s | Reading documents |
| Llama 3.2 1B Instruct | 8-bit | 128K | 214 tok/s | Not rated |
| Qwen3 VL 2B Instruct | 8-bit | 40K | 124 tok/s | Text and images |
| Qwen2 VL 2B Instruct | 8-bit | 32K | 120 tok/s | Text and images |
| MiniCPM5 2B | 8-bit | 104K | 113 tok/s | Not rated |
| Gemma 3 4B IT | 8-bit | 112K | 75 tok/s | Text and images |
| Llama 3.2 3B Instruct | 8-bit | 32K | 85 tok/s | Not rated |
| LightOnOCR 2 1B | 8-bit | 16K | 235 tok/s | Text and images |
| Qwen3 1.7B Base | 8-bit | 32K | 150 tok/s | Not rated |
| HunyuanOCR | 8-bit | 60K | 222 tok/s | Text and images |
| Qwen3 0.6B Base | 8-bit | 32K | 350 tok/s | Not rated |
| Rax 4.5 | 8-bit | 256K | 118 tok/s | Text and images |
| Qwen2 VL 7B Instruct | 5-bit | 20K | 54 tok/s | Text and images |
| Ornith 1.5 9B | 4-bit | 32K | 56 tok/s | Not rated |
| Unlimited OCR 2 | 8-bit | 60K | 217 tok/s | Reading documents |
| Granite 4.1 3B | 8-bit | 44K | 79 tok/s | Not rated |
| DeepSeek R1 Distill Qwen 1.5B | 8-bit | 128K | 165 tok/s | Not rated |
| Mistral 7B V0.1 | 6-bit | 32K | 51 tok/s | Not rated |
| Qwen2.5 Coder 3B Instruct | 8-bit | 32K | 87 tok/s | Coding |
| Meta Llama 3 8B Instruct Abliterated V3 | 5-bit | 8K | 56 tok/s | Not rated |
| EXAONE 3.5 7.8B Instruct | 4-bit | 16K | 66 tok/s | Not rated |
| Saiga Llama3 8B | 5-bit | 8K | 56 tok/s | Not rated |
| MiniCPM V 4.7 1B | 8-bit | 256K | 193 tok/s | Text and images |
| Qwen2.5 Coder 1.5B Instruct | 8-bit | 32K | 165 tok/s | Coding |
| DeepSeek R1 0528 Qwen3 8B | 3-bit | 20K | 78 tok/s | Not rated |
| Qwen3 4B Thinking 2507 | 8-bit | 16K | 68 tok/s | Not rated |
| Phi 4 Mini Instruct | 8-bit | 20K | 72 tok/s | Not rated |
| Dots.ocr | 8-bit | 128K | 97 tok/s | Reading documents |
| DeepSeek R1 Distill Qwen 7B | 6-bit | 16K | 52 tok/s | Not rated |
| Phi 3 Mini 4K Instruct | 8-bit | 4K | 74 tok/s | Not rated |
| DeepSeek OCR 2 | 8-bit | 8K | 210 tok/s | Reading documents |
| SmolLM3 3B | 8-bit | 52K | 87 tok/s | Not rated |
| Qwen2 7B Instruct | 6-bit | 16K | 52 tok/s | Not rated |
| Apertus 8B Instruct 2509 | 4-bit | 16K | 65 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 234 are shown.
Image and video models
| Model | Best fit | Per output | Strongest at |
|---|---|---|---|
| Stable Diffusion XL 1.0 | 16-bit | 16 s | Images |
| Z-Image Turbo | 8-bit | 23 s | Images |
| FLUX.1 [dev] | 4-bit | 52 s | Images |
| FLUX.3 Klein | 8-bit | 33 s | Images |
| Stable Diffusion 3.5 Large | 5-bit | 3.7 min | Images |
| Wan 2.2 TI2V-5B | 8-bit | 34 min | Video |
Only just too big
- Ministral 3 14B
- Qwen3 14B
- Qwen2.5 Coder 14B Instruct
- Qwen2.5 14B Instruct
- Gemma 3 12B IT
- Phi 3.5 Mini Instruct
- Qwen-Image 2.1
- Phi 4
- Phi 3 Mini 128K Instruct
- Pixtral 12B
- DeepSeek R1 Distill Qwen 14B
- Qwen2.5 14B Instruct 1M
- FLUX.2 [klein] 9B
- CodeLlama 7B HF
- DeepSeek V2 Lite
- Llama Guard 4 12B
- Gemma 2 9B IT
- DeepSeek Coder V2 Lite Instruct
- Qwen1.5 7B
- Krea 2 Turbo
- Deepseek Coder 6.7B Base
- Kimi VL A3B Instruct
- Llama 3.2 11B Vision Instruct
- Qwen 7B Chat
- Qwen1.5 MOE A2.7B
- Llama2 13B Chat HF
- Midm 2.0 Base Instruct
- Phi 4 Reasoning
- HarmBench Llama 2 13B Cls
- Moonlight 16B A3B Instruct
These need up to a quarter more memory than the RTX 3070 has. Spilling into system RAM or a lower precision can get them running, slowly.
Questions
What is the best LLM for the RTX 3070?
Of the open models we track, the strongest that run well on the RTX 3070 are Qwen3.5 9B, Gemma 4 12B, Qwen3.5 4B. 234 text models fit it in all.
Calculated from published specs, not benchmarked. All machines