Browse · topic

Telegram channels by topic

LLM classification, not a measurement. A language model reads a channel’s own title, description and recent posts and assigns the closest of 31 fixed categories, with a confidence score. A subscriber count is read off a public page; a topic is a judgement, and this one can be wrong about any channel listed. Treat a topic as the model’s reading of what a channel is likely about, nothing more. Within every list, channels are ordered by their measured subscriber count — the topic decides who is on the list, not where they sit on it.

Coverage, as read. The classifier has read 3,215,184 of the 4,183,741 entries the register holds a subscriber measurement for — 76.8% of them. Its most recent classification was written on 27 October 2026. 6 model/prompt-version combinations have contributed classifications so far. The figures below were measured at 27 Oct 2026, 13:08 UTC.

Topics, largest first

Other / unclassifiable is last and has no page. It is where the classifier puts an entry it could not place under any other category, so it is the absence of a topic rather than a topic. It is counted here so the table adds up to every entry the classifier has read. A topic with fewer than 200 entries would be counted here but not given a page. Membership and order on each topic’s page come from a ranking rebuilt hourly, and the page itself is cached for up to an hour more, so a topic page can trail the live register by a couple of hours.

Other ways in

The same register, arranged by measurement instead: subscriber size, the language of a channel’s own posts, or how recently it last published. How we measure.