Open vs Closed LLMs

Companies often prefer open-source LLMs for e.g. privacy, cost, security. This unit helps you understand what that means and why.

LLM models are fundamentally made of “weights”, typically billions to trillions of them. Each weight is really just a number, e.g. “1.12345”. Literally. Sounds magical right?

For practical understanding, you do not need to understand exactly why having billions of numbers enable LLMs to speak English. For simplification, you can think of each LLM as simply the collection of these billions of weights. Each LLM has a different weight (different set of billions of numbers).

Analogy

LLM weights are like a restaurant recipe 👨‍🍳.

Closed-source model: only the company that developed the recipe has access to the recipe. Nobody else does (and this makes some sense because they literally spend billions to develop that “recipe”). So OpenAI is the ONLY restaurant that serves this food item. You and I can USE the model (order this food item), but we cannot cook this food item ourselves.

Open-source models: someone (usually a company) developed the recipe (weights) and make them public. Everyone can see the recipe (download the weights), and thus cook the food and sell it. Now you might ask: why the heck would anyone do that and give it out for free? That’s a good question, and this is a good question. See ChatGPT's answer.

Today, most popular closed source models are GPT series (OpenAI), Claude (Anthropic), Gemini (Google). Popular open source models include Deepseek (Deepseek), Qwen (Alibaba).

Choosing Open vs Close

Why one might prefer open source: Open source LLMs allow your organization to run the LLM on their own servers (same idea as “on-premises”). This way you do not need to send sensitive data to e.g. OpenAI.

Why one might prefer close source: (I might trigger open source model advocates with this (ᵕ—ᴗ—) ) As of 2026, the frontier closed source models are IN GENERAL more intelligent than open source models. Although some people argue the gap is reducing. See Artificial Analysis Intelligence Index of open vs. closed source LLMs: https://artificialanalysis.ai/models#intelligence-category-tabs