How much memory popular open-weight models need at common quantization levels, and the cheapest hardware in our dataset that fits each one. Parameter counts come from the official model cards; memory figures are estimates using the method on our methodology page.
6 articles
Can I Run DeepSeek Locally? Hardware Requirements
Hardware needed to run DeepSeek-R1 and its distilled models locally, from 8B to the full 671B, with memory…
Read →How to Run Qwen3 Locally
Memory and hardware needed to run Qwen3 models locally, from 4B to 235B-A22B, including the mixture-of-experts…
Read →How Much VRAM for Llama? 8B, 70B and 405B
VRAM needed to run Llama 3.1 and 3.3 locally at different quantizations, with the cheapest GPU or system that…
Read →gpt-oss Hardware Requirements
What you need to run OpenAI gpt-oss-20b and gpt-oss-120b locally, using OpenAI's published memory…
Read →Gemma 3 VRAM Requirements
VRAM needed to run Google Gemma 3 models locally at Q4 and Q8, and which GPUs…
Read →Running 70B Models at Home
What it takes to run a 70B-parameter LLM at home: dual GPUs, 128 GB unified-memory mini PCs or a Mac…
Read →