Llama

Open source AI vs proprietary AI compared: what licenses allow, how far the capability gap has narrowed, and when self-hosting actually beats a paid API.

Running an open-source LLM locally — on your laptop or home server — appeals to privacy-conscious users and developers. No data sent to OpenAI or Anthropic. Full control. The marketing is seductive. But the realistic cost is high, and most people who start down this path abandon it. A capable local LLM model (Llama 3, Mistral 8B, Deepseek) requires 16–48 GB of RAM or VRAM, quantization to run at reasonable speeds (which degrades quality), hours of generation per response compared to seconds on cloud APIs, significant electricity consumption, heat management, and battery drain on laptops. When you sum the hardware cost, electricity, and opportunity cost of slower responses, running a local model often costs more than a cloud API — even when you factor in privacy. The honest assessment: local LLMs make sense for specific use cases (offline work, sensitive research, air-gapped systems). For most knowledge workers, cloud APIs remain cheaper and faster.

Llama 4 is Meta's April 2025 open AI family — Scout, Maverick, and the shelved Behemoth — with native multimodal input and a 10M-token context window.

Run open-source AI on your computer with Ollama. Llama 3.1, Mistral, and Phi compared, plus real hardware requirements and setup steps.