Open-Source vs. Closed AI Models — and Why It Matters

Open-source (or 'open-weight') models like Meta's Llama, Mistral, and DeepSeek can be downloaded, modified, and run on your own computer or server. Closed models like GPT and Claude can only be accessed through the company's API. Open lets you customize and avoid lock-in; closed gives you the smartest models, easier setup, and built-in safety.

Some examples

• Llama 3.1 70B runs on a $3,000 gaming PC — slowly, but it works, and your data never leaves your house.
• DeepSeek R1 (open-weight, late 2024) was competitive with GPT-4 on reasoning tasks at a fraction of the training cost.
• Ollama is a free Mac/Windows app that lets you download and chat with open-source models in 10 minutes.
• Closed models (GPT-5, Claude Opus 4.7) tend to be ~6 months ahead of open models on the hardest benchmarks.

Try it!

Download Ollama (ollama.com) and run 'ollama run llama3.2' in your terminal. In about 5 minutes, you have a real AI model running locally with no API key, no account, no data leaving your laptop. Free forever.