Meta’s Muse Wonder Full Model Rivals Top AI Systems in New Benchmarks

Meta has released Muse Wonder and Muse Wonder Full, asserting that these new artificial intelligence models place the company back at the forefront of global AI development. According to Wired, the launch signals that Mark Zuckerberg’s recent massive investment in AI research is yielding competitive results against industry leaders.

The company claims Muse Wonder Full possesses frontier capabilities, positioning it among the most powerful models available today. Meta stated in a blog post that these releases represent a significant leap toward personal superintelligence. The models are designed to compete directly with the highest-performing systems from rival firms.

Context and Background

Meta announced the first model from its newly formed Meta Superintelligence Labs, Muse Spark, in April. The company also released a smaller open-weight model called Muse Glimmer and indicated plans for larger open-weight releases later. However, Meta did not disclose specific details regarding how the new Muse Wonder models were built or trained.

Zuckerberg overhauled the company’s AI strategy last year by investing billions into Meta Superintelligence Labs. He hired Alexandr Wang, the former CEO of Scale AI, to lead the division. This restructuring has faced challenges, with some researchers leaving the organization shortly after joining, according to reports.

Meta was previously a leader in open-weight AI with its Llama series, which allowed users to download and run models on their own hardware. The company fell behind after releasing several underperforming versions while Anthropic, OpenAI, and Google introduced more capable alternatives. The new Muse Wonder models aim to reverse that trend.

Who Is Affected

Developers and enterprises can access the Muse Wonder models through an API and via Muse Code, a software development tool. Meta says the best version of the model will power features across its platforms, including its Muse personal agent. This integration suggests a shift toward embedding advanced AI capabilities directly into consumer-facing services.

On the Artificial Analysis Intelligence Index, Meta says the largest Muse Wonder model outperforms Anthropic’s second-best model, Claude Fable 5.1. It ranks just below Anthropic’s top model, Claude Opus 5.5. This aggregated score is widely used to measure model capabilities across various tasks.

Meta further claims the model beats OpenAI’s best system, GPT-6 Astra, on specialized benchmarks. These specific tests include measuring a model’s ability to work with legal documents and to build software. The company did not provide independent verification of these benchmark results in the source material.

What should you do?

Security and technology professionals should monitor the release of new AI models for changes in threat landscapes, as more capable systems can be used for sophisticated social engineering or code generation. Organizations using AI APIs should review their access controls and data handling policies to ensure sensitive information is not inadvertently exposed to third-party models.

Background: Social engineering

Look for urgency, secrecy, and deviation from standard channels. Verify identity through a separate, known communication path before acting. Isolate suspicious interactions immediately to prevent lateral movement within your network environment.

Read the full guide: Social Engineering Warning Signs and How to Respond

What to do and how to stay safe: Meta

• Review API access logs for any unusual activity following the integration of new AI models into your infrastructure.
• Verify that data sent to external AI services does not contain sensitive or proprietary information that could be exposed.
• Monitor vendor announcements for security patches or updates related to the new model versions as they become available.
• Assess current incident response plans to ensure they cover threats involving advanced AI-generated content or automated attacks.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

How does Muse Wonder compare to other leading AI models?

Meta states that Muse Wonder Full outperforms OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1, while ranking just below Claude Opus 5.5 on the Artificial Analysis Intelligence Index.

Is the Muse Wonder model open-weight?

Meta has not disclosed details on how Muse Wonder was built, but it plans to release larger open-weight models in the future, following the recent release of the smaller Muse Glimmer.

Who is leading Meta’s new AI research efforts?

Alexandr Wang, the former CEO of Scale AI, leads Meta Superintelligence Labs, which was formed last year after Mark Zuckerberg overhauled the company’s AI strategy.