What AI model does Telegram use?
Telegram's AI features run on Cocoon, a secure, open-source, decentralised AI platform that Telegram built for exactly this purpose. The question of which specific model is secondary to the question of where it runs, and that is the part Cocoon answers.
Why the platform matters more than the model name
Adding AI to a messenger usually means forwarding user text to an external provider, at which point the provider sees the content and the messenger's privacy promise stops at its supplier's door. Cocoon exists so that does not happen: processing takes place in an isolated, encrypted environment, distributed rather than concentrated in one company's data centre, with the implementation published as open source so it can be inspected rather than believed.
Which features use it
- Translation
- Custom tones
- AI summaries of long channel posts and Instant View articles
- AI-powered sticker search
Third-party AI is a different thing entirely
If you use AI through a bot or a mini app, you are using that developer's service. Whatever model they chose, whatever provider they send your text to, and whatever policy they operate under all apply instead. Telegram's arrangement does not extend to them, and the interface will not tell you apart, so check who runs a bot before sending it anything you would not want a stranger to hold.