C EN Meta Superintelligence Labs·Release·

Introducing Muse Wonder and Muse Wonder Full

C · Weak signal Meta newsroom launch post that claims state-of-the-art agentic performance without publishing a single benchmark, price or evaluation; the safety report is promised for later.

Summary based on the full text·Original: about 3 minutes to read

In short

Meta's announcement (12 October 2026) of Muse Wonder and Muse Wonder Full, which it calls its most capable models, built on its earlier Muse Spark. Muse Wonder is available to developers through the Meta Model API and Muse Code; Muse Wonder Full is limited to Muse Code subscribers. The post names two features, Active Memory and Active Focus, and lists safety intentions. It gives no numbers on performance.

Key numbers

  • Fourteen monthstime since Meta began overhauling its AI efforts, per the post

As reported by the source.

Read the original on about.fb.com

English·Web page·opens about.fb.com

What it reports

  • Two models, positioned for personal and professional agent tasks in law, finance, health and coding.
  • Active Memory carries details across conversations; Active Focus keeps working on hard tasks, recovers from errors and asks for clarification before important steps.
  • Safety: trained to avoid sharing private information without permission, refuse harmful requests and support defensive security work, under Meta's Superintelligence Scaling Framework.
  • A full safety evaluation report is "available soon".
  • Muse Wonder will later power the consumer Muse product, first for subscribers.

How they got it

A product announcement. No benchmarks, methodology or comparisons appear in the post; it points to a separate research blog for detail.

Limits

  • No benchmark scores, prices or context lengths in the post; claims of state-of-the-art performance cannot be checked from it.
  • The safety report was not published at launch.
  • Pricing and independent index scores circulated in press coverage, not in this source.

Why it matters for companies

Meta now has a frontier-class model on a paid API, which adds a vendor to procurement shortlists. Until independent evaluations and the safety report are out, treat it as a candidate to test, not a replacement. Ask for data-handling terms around Active Memory before any pilot that touches personal or customer data.

Summary based on the full text. Read the original for the full argument and data.Read the original on about.fb.com

Tags

Same week

All of Week 42