Meta has launched Muse Wonder, an advanced AI model internally known as Watermelon, that reportedly rivals the capabilities of ChatGPT and Claude in coding and professional domains such as finance and legal work. However, Meta has not clarified whether it will publish the model's weights, a key factor under the European Union's AI Act.
- Muse Wonder matches ChatGPT and Claude on coding and professional work.
- Meta has not confirmed publishing model weights, impacting EU AI Act obligations.
- Model accessible through Meta Model API and Muse Code, powering paid services.
What happened
Meta has released Muse Wonder, also known as Watermelon internally, as the most capable AI model from its revamped AI division. The company says Muse Wonder matches the coding and professional task performance of top AI models like ChatGPT and Claude, despite not publishing formal benchmarks. The model is currently accessible through the Meta Model API and Muse Code, and it powers the paid tier of Meta’s Muse agent launched recently.
This release follows Meta's previous frontier claims earlier in the year and significant investments by Mark Zuckerberg, including a $14 billion deal for a stake in Scale AI to support the AI team. Meta’s announcements come after meeting its internal timeline set in August for launch by October, although the company has remained cautious about revealing detailed model attributes publicly.
Why it matters
Muse Wonder’s claimed parity with ChatGPT and Claude positions Meta strongly in the competitive AI landscape, especially given the high-profile investments and restructures behind the model. However, transparency around the model weights, which determine regulatory responsibilities under the European Union’s AI Act, is a critical unresolved issue. Meta has not confirmed if it will publish these weights, which would affect the level of regulatory scrutiny the model faces.
Under the EU AI Act, publicly released model weights can exempt AI systems from certain documentation demands but only if the license is truly open and comprehensive. Models posing systemic risks must comply fully regardless of openness, and Meta has so far refrained from signing the relevant code of practice. This regulatory grey area could influence Meta’s ability to deploy Muse Wonder widely within Europe.
What to watch next
Observers should monitor whether Meta decides to publish Muse Wonder’s model weights, as this will clarify the company’s compliance posture under evolving European regulations. Meta’s history of cautious openness around flagship AI systems hints that full transparency may not be forthcoming, but pressure from regulators and market participants may change the stance.
The performance and adoption of Muse Wonder via developer APIs and in Meta’s paid AI services will also provide insight into the model’s practical impact and user reception. Additionally, further product releases or updates that address issues such as privacy settings or operational risks—as seen with the quick shutdown of Muse Image—will be important indicators of how Meta balances innovation with responsibility.