Meta’s Muse Wonder pairs a million-token context with subscriptions

Built for work that carries on

Active Memory is designed to retain useful details between conversations and bring them back when needed. Muse Wonder can also work behind a laptop screen, plan longer tasks, hand parts to subagents and recover from errors. The account describes an approach similar to Anthropic’s Claude: give the model access to resources, then let it work.

That promise comes as Meta tries to close ground in AI. Thomas Smith of Fast Company says the company had long appeared to be lagging, but that Muse began taking share from competitors soon after launch. In less than a month, the service reached 6.6 million downloads and 1.8 million daily users. Citi estimates it could generate $27 billion in revenue relatively quickly.

The benchmark picture

On the Artificial Analysis Intelligence Index, Muse Wonder Full scored 56, matching Claude Sonnet 5.5 and beating GPT-6 Astra, Gemini 4 Argon and Claude Fable 5.1. The standard Muse Wonder scored 52. The article also points readers to the models’ SWE-bench results.

The score gives Meta a strong comparison point, but it does not settle what users can make with the model. Muse Wonder’s output is written plans and documents: it can take varied inputs, but it does not generate video or images. That may suit paralegals; it is a poor fit for someone who needs an infographic or a video.

The video-generation services named in the article are Runway, Synthesia, Pictory AI, Klap and Descript. With OpenAI Sora gone, the author sees no clear leader in the category.

The product boundary

Muse Wonder costs $4 per million input tokens and $20 per million output tokens. Muse Wonder Full is available only on the highest subscription tier.

I think the important distinction is not the benchmark tie but the gap between what the agent can do and what it can produce. Meta is making a case for persistent, multi-step work, yet the deliverable remains text. If the company’s new models are to hold a place in the market, their value will depend on whether that written output is enough for the people Meta is trying to attract.