Open-source vs. closed AI models: what's the difference

When people talk about large language models, they're often split into "open" (open-source or open-weight) and "closed" (proprietary) categories. The difference isn't just about the price of access — it's about what exactly a company discloses to the public, and who ultimately controls the model.

What an "Open" Model Means

An open model has publicly available weights — the trained numerical parameters of the neural network — which can be downloaded and run on your own hardware without going through the developer's API. This doesn't always mean the full training source code or training data is open too — more often, it's just the finished model weights that get released, not the entire process behind creating them.

What a "Closed" Model Means

A closed model's weights aren't publicly available — users interact with it only through the company's API or interface, and the company fully controls how the model runs, gets updated, and gets moderated, and can change access terms or shut the service down at any time.

Pros and Cons of Each Approach

• Open models offer independence from a single vendor, the ability to fine-tune the model for your specific task, and the option to run it locally without sending data to a third party — but they require your own compute and expertise to deploy
• Closed models are usually easier to use and often lead open alternatives in quality at launch — but they tie users to a specific vendor and its access and pricing policy

What This Means in Practice

Choosing between an open and a closed model is mainly a choice between control and convenience: an open model gives more independence at the cost of deployment complexity, while a closed model gives ease of use at the cost of depending on a single vendor. Many companies and developers use both approaches at once for different tasks, rather than picking one solution once and for all.