Skip to content
Vector Tribune
Security news, verified and fast
EU AI Act compliance hangs on whether Meta releases Muse Wonder weights after matching rivals
AI

EU AI Act compliance hangs on whether Meta releases Muse Wonder weights after matching rivals

Meta’s new Muse Wonder model matches competitors in coding tasks, but lacks public benchmark data and open weights.

Key points

  • Muse Wonder matches ChatGPT and Claude on coding and professional work.
  • Meta has not published benchmarks or confirmed if model weights are open.
  • EU AI Act compliance depends on whether Meta releases the underlying weights.

Meta has released Muse Wonder, the most capable model from its rebuilt AI unit. Internally known as Watermelon, the model reached its target release window in October. Chief AI officer Alexandr Wang told Bloomberg the system matches ChatGPT and Claude on coding and professional tasks.

The risk

The claim of parity arrives without published benchmarks. Wang argued that benchmarks are weak proof because they can be gamed. He suggested developers try the model themselves to judge its quality. This follows similar claims for Muse Spark 1.3 in September and the first Muse Spark in April.

Who is exposed

Developers can access Muse Wonder through the Meta Model API and Muse Code. The model will also power the paid tier of the Muse agent launched last month. The previous Muse Spark model scored fourth on one index, trailing leaders on reasoning benchmarks.

Cost and compliance

Zuckerberg committed hundreds of billions to infrastructure and paid $14B for a Scale AI stake. The key question for Europe is whether Meta publishes the weights. Under Article 53 of the EU AI Act, free and open models escape some documentation duties. Meta has not decided on Muse Wonder’s licensing status.

What should you do?

Managers should verify if the model meets their specific coding or legal requirements through independent testing. Teams must assess whether the lack of open weights impacts their regulatory compliance strategy under the EU AI Act.

What to do and how to stay safe: Meta

  • Test the model against your specific coding and professional workloads to verify the parity claims.
  • Monitor Meta’s official announcements regarding the release of model weights and licensing terms.
  • Review your organization’s EU AI Act obligations for models that do not provide open weights.
  • Evaluate the risks of relying on a model without published benchmark scores for critical tasks.

Step-by-step guide: AI Governance Explained: Building Rules for Autonomous Systems

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

Frequently asked questions

Does Muse Wonder have published benchmarks?

No, Meta has not published benchmarks for Muse Wonder, though it claims parity with rivals.

Will Meta release the weights for Muse Wonder?

Meta has not yet decided whether to publish the weights for the Muse Wonder model.

How does this affect EU AI Act compliance?

If Meta does not publish the weights, the model may face stricter documentation requirements under Article 53.

Sources

  1. The Next Web
MetaMuse WonderEU AI ActAlexandr WangScale AI

Related stories