Meta Releases Muse Wonder and Muse Wonder Full: Two Points Short of Opus 5.5

Meta has released Muse Wonder and Muse Wonder Full, two models in a new flagship family. By the company's own account, it is Meta's strongest lineup to date. The target is stated outright: "personal superintelligence." According to Jiqizhixin, the flagship leads in several benchmarks and trails Anthropic's Opus 5.5 by just two points on the composite intelligence score.

It is a notable shift in Meta's position. Not long ago, the company was seen as playing catch-up.

What we know about the release

The models launched together, as a pair. Meta stresses Agent scenarios: both the user's personal tasks and professional work. The "personal assistant" language here describes a product goal, not a marketing add-on. Mark Zuckerberg set out this same framing back in 2025, when he wrote about personal superintelligence as a company priority.

The source this piece relies on cuts off at the very beginning, so some technical details are still unknown. Here is what can be stated:

• Family: Muse Wonder and Muse Wonder Full.
• Positioning: "personal superintelligence," Agent tasks at personal and professional levels.
• Meta's claim: the strongest model lineup in the company's history.
• Results: leads in a number of benchmarks, according to Jiqizhixin.
• Gap to Opus 5.5: two points on the intelligence score.
• Not disclosed in the available material: parameter count, context window, price per token, license, and whether the weights are open.

Why two points matter

Two points on a composite scale is a small gap, but the context makes it significant. In April 2025, Llama 4 got a lukewarm reception, and Meta had to rebuild its AI effort: Alexandr Wang joined the company as part of the Scale AI deal, a superintelligence lab appeared, and large hiring campaigns began. Muse Wonder is the first result of that overhaul to be compared with the market's leading edge rather than with earlier Llama versions.

There are several caveats. A composite index averages across different tasks, so on any specific job the gap could be larger or smaller. From the available text, it is unclear which scale the publication uses. It is also unclear which results Meta measured itself and which independent labs confirmed. The benchmark leadership the publication reports should be checked against independent measurements under the same conditions.

Open questions

The main question is openness. Llama was built on open weights, and that was Meta's main selling point with developers. Whether Muse Wonder will follow the same path or become a closed model is not clear from the material. The answer determines who can deploy it on their own infrastructure and who will pay for the API.

The second question is distribution. Meta's family of apps has billions of daily users, a resource that neither Anthropic nor most other competitors have. If Muse Wonder is built into Meta AI, the messengers, and smart glasses, a personal agent would get an audience without a separate acquisition campaign.

The third question is price. A model two points behind the leader can still win if it costs noticeably less.

What developers and teams should do

It is too early to rewrite your stack over two points. It makes more sense to wait for independent measurements and, until then, run Muse Wonder on your own tasks: tool calling, long chains of actions, and code work. Agent scenarios most often break at the third to fifth step, and a composite index does not show that.

Watch three things. First, access terms: API, license, regional restrictions. Second, the cost of a full agent run, not a single request. Third, the stability of responses on your own test set. If all three turn out to be acceptable, Muse Wonder is worth adding to a multi-model setup as a backup or cheaper option alongside Opus.