What are open weights?

After training, a model is essentially a very large set of internal numeric settings that determine how it responds. When a company publishes those settings as downloadable files, anyone can run that model on their own hardware or through a third-party host, that's what 'open weights' means. A closed model, by contrast, is only usable through the company's own interface or paid access. The term isn't identical to 'open source': the training data and the exact process used to build the model usually aren't included, only the finished settings.

This matters for AI visibility because open-weight models get picked up and redeployed by many different companies, each wrapping it with its own system prompt, tools, and rules, so the same base model can behave very differently depending on who's running it. It also matters for crawling: plenty of smaller AI products and aggregators run on an open-weight model without operating their own named, identifiable crawler, so robots.txt rules written against recognizable bots like GPTBot or ClaudeBot won't necessarily have any effect on how those products absorb or reach a site's content.