Best local LLMs for 10 GB of VRAM
Every open model we track that runs fully in 10 GB of GPU memory, biggest first. Checks are at Q4_K_M with about 4k context.
- Run fully at Q4
- 340 of 656
- Run at Q8 too
- 198
- Biggest at Q4
- HyperCLOVAX-SEED-Omni-8B
With 10 GB, the biggest model you can run comfortably is HyperCLOVAX-SEED-Omni-8B (~8.5 GB at Q4). If you want higher quality at Q8, YuE-s1-7B-anneal-en-cot is the biggest that fits.
Models that fit in 10 GB
Bar: Q4 memory needed vs. 10 GB. The tick marks 90%, our comfort line.FitsTight
| Model | Q4 need vs. 10 GB | Q4 / Q8 |
|---|---|---|
| Ornith-1.5-9B9.7B params, ornith-ai | 7.7 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen3.5-9B9.7B params, Qwen | 7.7 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen3-VL-8B-Instruct8.8B params, Qwen | 7.1 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen2.5-VL-7B-Instruct8.3B params, Qwen | 6.8 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen3-8B8.2B params, Qwen | 6.7 / 10.0 GB | Q4FitsQ8Won’t fit |
| granite-3.3-8b-instruct8.2B params, ibm-granite | 6.7 / 10.0 GB | Q4FitsQ8Won’t fit |
| Llama-3.1-8B-Instruct8.0B params, meta-llama | 6.6 / 10.0 GB | Q4FitsQ8Won’t fit |
| Llama-3.1-8B8.0B params, meta-llama | 6.6 / 10.0 GB | Q4FitsQ8Won’t fit |
| Nemotron 3.5 Bolt E4B8.0B params, nvidia | 6.6 / 10.0 GB | Q4FitsQ8Won’t fit |
| Gemma 4 E4B8.0B params, google | 6.6 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen2.5-Coder-7B-Instruct7.6B params, Qwen | 6.3 / 10.0 GB | Q4FitsQ8Won’t fit |
| DeepSeek-R1-Distill-Qwen-7B7.6B params, deepseek-ai | 6.3 / 10.0 GB | Q4FitsQ8Won’t fit |
| Qwen2.5-7B-Instruct7.6B params, Qwen | 6.3 / 10.0 GB | Q4FitsQ8Won’t fit |
| Mistral-7B-Instruct-v0.37.2B params, mistralai | 6.1 / 10.0 GB | Q4FitsQ8Tight |
| Gemma 4 E2B5.1B params, google | 4.6 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3.5-4B4.7B params, Qwen | 4.2 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3-VL-4B-Instruct4.4B params, Qwen | 4.1 / 10.0 GB | Q4FitsQ8Fits |
| gemma-3-4b-it4.3B params, google | 4.0 / 10.0 GB | Q4FitsQ8Fits |
| Spark-X2.6-4B4.1B params, XHToken | 3.9 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3-4B4.0B params, Qwen | 3.8 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3-4B-Instruct-25074.0B params, Qwen | 3.8 / 10.0 GB | Q4FitsQ8Fits |
| Nemotron Labs Dory 1.0 4B4.0B params, nvidia | 3.8 / 10.0 GB | Q4FitsQ8Fits |
| Phi-4-mini-instruct3.8B params, microsoft | 3.7 / 10.0 GB | Q4FitsQ8Fits |
| Llama-3.2-3B-Instruct3.2B params, meta-llama | 3.2 / 10.0 GB | Q4FitsQ8Fits |
| Qwen2.5-3B-Instruct3.1B params, Qwen | 3.2 / 10.0 GB | Q4FitsQ8Fits |
| MiniCPM5-2B2.5B params, openbmb | 2.8 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3.5-2B2.3B params, Qwen | 2.6 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3-1.7B2.0B params, Qwen | 2.4 / 10.0 GB | Q4FitsQ8Fits |
| DeepSeek-R1-Distill-Qwen-1.5B1.8B params, deepseek-ai | 2.2 / 10.0 GB | Q4FitsQ8Fits |
| LFM3-1.7B1.7B params, LiquidAI | 2.2 / 10.0 GB | Q4FitsQ8Fits |
| Qwen2.5-1.5B-Instruct1.5B params, Qwen | 2.1 / 10.0 GB | Q4FitsQ8Fits |
| Llama-3.2-1B-Instruct1.2B params, meta-llama | 1.9 / 10.0 GB | Q4FitsQ8Fits |
| gemma-3-1b-it1.0B params, google | 1.7 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3.5-0.8B0.9B params, Qwen | 1.6 / 10.0 GB | Q4FitsQ8Fits |
| Qwen3-0.6B0.8B params, Qwen | 1.5 / 10.0 GB | Q4FitsQ8Fits |
| Qwen2.5-0.5B-Instruct0.5B params, Qwen | 1.3 / 10.0 GB | Q4FitsQ8Fits |
| Mistral-Nemo-Instruct-240712.2B params, mistralai | 9.5 / 10.0 GB | Q4TightQ8Won’t fit |
| gemma-3-12b-it12.2B params, google | 9.5 / 10.0 GB | Q4TightQ8Won’t fit |
| Mellum2.1-12B-A2.5B-Thinking12.1B params, JetBrains | 9.5 / 10.0 GB | Q4TightQ8Won’t fit |
| Gemma 4 12B12.0B params, google | 9.3 / 10.0 GB | Q4TightQ8Won’t fit |
Hardware we track with 10–11 GB usable
- Intel Arc B57010 GB usable, $219
- GeForce RTX 3080 10GB10 GB usable, $699
- GeForce RTX 2080 Ti11 GB usable, $999
Quick answers
- What is the biggest LLM I can run with 10 GB of VRAM?
- HyperCLOVAX-SEED-Omni-8B (10.7B parameters) is the biggest model we track that runs comfortably in 10 GB at Q4_K_M, needing about 8.5 GB with a ~4k context. At Q8 the biggest is YuE-s1-7B-anneal-en-cot.
- How many models fit in 10 GB of VRAM?
- 340 of the 656 open models we track run fully in 10 GB at Q4 with room to spare, 198 also fit at Q8, and 9 more fit only with a short context.
- Can I run a model that needs more than 10 GB?
- Yes, with llama.cpp or Ollama you can offload some layers to system RAM, but speed drops sharply. For full-speed local inference the whole model plus its context cache should fit in GPU memory; we use 90% of 10 GB as the comfort line.
- Which hardware has about 10 GB of usable memory?
- The cheapest current device we track in this range is the Intel Arc B570 (10 GB usable, $219 list price as of our last snapshot). 3 tracked devices have 10–11 GB usable.
Other memory sizes
4 GB · 6 GB · 8 GB · 12 GB · 16 GB · 20 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB