Particle.news

Meta Unveils Muse Wonder Models to Power Paid Agent Services

Muse Wonder adds long‑term memory, a one‑million‑token context window, paid tiering, raising new security and enterprise trust questions.

Overview

  • Meta introduced Muse Wonder and Muse Wonder Full on Monday, October 12, and has begun moving Wonder into paid Muse assistant tiers while limiting Wonder Full to higher Muse Code subscriptions.
  • The models add Active Memory and Active Focus to carry context across conversations and to keep working on multi‑step tasks, and both support multimodal inputs and a one‑million‑token context window.
  • Meta published developer pricing for the Meta Model API and Muse Code, with Wonder priced at $4 per million input tokens and Wonder Full restricted to $50 and $200 Muse Code plans.
  • Early independent tests show competitive but mixed results versus Anthropic and OpenAI, and separate reports say serious security bugs were discovered shortly before launch that could enable sandbox escapes and access to internal systems.
  • The release follows Meta’s large infrastructure and Scale AI investments and signals a shift from free models to closed, paid offerings, but enterprises remain cautious about data access, integration and trust.