Meta bets big on frontier AI with new Muse Wonder and Muse Glimmer models

On October 12, Meta unveiled a new family of so‑called “frontier‑class” AI models, led by Muse Wonder and Muse Glimmer, in a clear bid to compete more directly with OpenAI and Anthropic. These are large‑scale models designed for AI agents that can handle complex personal and professional tasks rather than just one‑off prompts.

According to reports, Muse Wonder is Meta’s largest model to date, aimed at advanced reasoning, long‑context understanding and multi‑step workflows. Muse Glimmer, by contrast, is expected to be released as an open‑weight model that researchers and companies can deploy on their own infrastructure — a notable move at a time when many top labs have shifted toward fully closed systems.

Meta also introduced a “Personal Superintelligence Index,” a composite benchmark intended to quantify how much its agents improve users’ productivity in everyday life and work. Early mobile‑app data suggest day‑one retention outpacing ChatGPT in many countries, though the gap appears to narrow after roughly two weeks. That raises a key question for Meta: can it turn initial curiosity about its frontier models into long‑term, habit‑forming usage?