RunLLMLocal
Open the checker

Models

VRAM requirements for popular local LLMs

How much memory each open model needs, which quantization to use, and the cheapest GPU that runs it fully. 39 models from Qwen, Meta, Google, Mistral, DeepSeek, Microsoft and OpenAI.

ModelSizeQ4 at 8KCheapest GPU
Qwen3 0.6B0.6B0.6 GBGTX 1650
Gemma 3 1B1B0.9 GBGTX 1650
Llama 3.2 1B1.2B1.0 GBGTX 1650
Llama 3.2 3B3.2B3.0 GBGTX 1650
Phi-4 Mini 3.8B3.8B3.2 GBGTX 1650
Qwen3 4B4B3.7 GBGTX 1660 / 1660 Super
Gemma 3 4B4.3B4.7 GBGTX 1660 / 1660 Super
Qwen3.5 4B4.7B3.9 GBGTX 1660 / 1660 Super
Mistral 7B v0.37.2B5.7 GBRTX 3050
Qwen2.5-Coder 7B7.6B5.8 GBRTX 3050
Llama 3.1 8B8B6.2 GBRTX 3050
DeepSeek-R1 Distill 8B8B6.2 GBRTX 3050
Qwen3 8B8.2B6.4 GBRTX 3050
Qwen3.5 9B9.4B7.0 GBRTX 3050
Gemma 4 12B12B11 GBRTX 3060
Gemma 3 12B12.2B11 GBRTX 4060 Ti
Mistral Nemo 12B12.2B9.2 GBGTX 1080 Ti
Phi-4 14B14.7B11 GBRTX 4060 Ti
Qwen3 14B14.8B11 GBRTX 4060 Ti
DeepSeek-R1 Distill 14B14.8B11 GBRTX 4060 Ti
gpt-oss 20B21B MoE14 GBRTX 4060 Ti
Codestral 22B22.2B16 GBRX 7900 XT
Mistral Small 3 24B24B18 GBRX 7900 XT
Devstral Small 24B24B17 GBRX 7900 XT
Gemma 4 26B-A4B (MoE)25.8B MoE19 GBRTX 3090 / 3090 Ti
Gemma 3 27B27.4B22 GBRTX 5090
Qwen3.6 27B27.5B19 GBRTX 3090 / 3090 Ti
Qwen3 30B-A3B (MoE)30.5B MoE21 GBRTX 3090 / 3090 Ti
Qwen3-Coder 30B-A3B30.5B MoE21 GBRTX 3090 / 3090 Ti
Gemma 4 31B31B25 GBRTX 5090
Qwen3 32B32.8B23 GBRTX 5090
DeepSeek-R1 Distill 32B32.8B23 GBRTX 5090
Qwen2.5-Coder 32B32.8B23 GBRTX 5090
Qwen3.6 35B-A3B (MoE)35B MoE23 GBRTX 5090
Llama 3.3 70B70.6B48 GBMulti-GPU or Mac
DeepSeek-R1 Distill 70B70.6B48 GBMulti-GPU or Mac
Llama 4 Scout 109B-A17B109B MoE73 GBMulti-GPU or Mac
gpt-oss 120B117B MoE77 GBMulti-GPU or Mac
Qwen3 235B-A22B (MoE)235B MoE155 GBMulti-GPU or Mac
Copied