Meta Unveils Muse Wonder Models as Key to Unlocking Personal Superintelligence

Introduction

Meta has spent over a year rebuilding its AI operation around the concept of personal superintelligence. The company has now introduced the Muse Wonder series, a pair of models designed to handle complex workflows and agentic tasks.

What Muse Wonder Models Do

The Muse Wonder models are built for agentic tasks, both personal and professional settings. Two key features, Active Memory and Active Focus, are designed to handle complex workflows over long periods.

Background and Leadership

The team behind the Muse Wonder models includes leadership from outside the company. Alexandr Wang, previously of Scale AI, serves as Chief AI Officer, while Nat Friedman, formerly of GitHub, is the product lead.

Availability and Benchmarking

The Muse Wonder models have been benchmarked as state-of-the-art in three areas: memory, coding, and agent performance. They are available immediately through the Meta Model API and Muse Code, with the Full version reserved for subscribers.

Training and Distribution

Training took place at Prometheus, Meta's gigawatt-scale data center. The team behind the model has grown to around 3,000 to 3,400 staff, including high-profile hires from rivals such as OpenAI and Google DeepMind.

Implications for Meta and the AI Race

The distribution of the Muse Wonder models is Meta's quiet advantage. The company describes the models as aimed at personal superintelligence, with the potential to serve billions of users. A paid AI tier for Muse Wonder Full would mark a meaningful addition to how the company monetizes its products.