About

It's hard to tell from a single tweet how significant an observation of a model's behavior is: how much of the output is explained by the user's prompt, and how much by the model's personality? This website aggregates dozens of reports per model to give a bird's-eye view of what people say about each LLM.

It is focused on model properties that are not capabilities, like character, personality traits, writing style, self-perception, or any weird quirk of the models.

How it works

  1. Every hour, new tweets from the Community Archive are scanned for mentions of current models.
  2. Each matching tweet is read together with the tweet it replies to, the tweet it quotes, its top replies and its attached media. Gemini 3.8 Flash decides whether it describes a behavior (not a capability, benchmark or piece of news) and scores how interesting it is from 1 to 10. Reports scoring 4 or more appear on the site.
  3. Reports are ranked by 50 × quotes + 20 × log2(1 + min(retweets, 100)) + 15 × interestingness, where quotes counts distinct Community Archive accounts that quote-tweeted the report, which is a sign of a report people actually discussed.
  4. Once a day, Claude Opus 5.5 reads each model's reports and updates the one-line trait and short portrait shown on its card.
  5. New releases on OpenRouter are checked daily and added automatically.

Credits

Made by Alexandre Variengien. The code is open source on GitHub. Data from the Community Archive.