Urgent.News

What's breaking now, across thousands of outlets.

AI

Meta launches Muse Wonder, catching OpenAI and Anthropic at the frontier

Meta has released a new AI model family called Muse Wonder. The new model leapfrogs Gemini 4, GPT-Astra, and Fable 5 in Artificial Analysis's leaderboard.

Meta launches Muse Wonder, catching OpenAI and Anthropic at the frontier

Meta unveiled a new family of AI models called Muse Wonder, which is now accessible to developers. This marks Meta's entry into the AI frontier, as indicated by independent benchmarks. The launch follows six weeks after the debut of Muse Spark 1.3, the latest iteration of a model family introduced in April 2023. Last year, Meta invested heavily in AI recruiting and infrastructure, culminating in the acquisition of Scale AI for $14.3 billion and the appointment of Alexandr Wang as its new Chief AI officer. These efforts seem to have yielded results.

Artificial Analysis, an independent AI benchmarking organization, found that Muse Wonder outperformed powerful models like GPT-Astra from OpenAI, Fable 5 from Anthropic, and the newly released Gemini 4 from Google. On the Artificial Analysis Intelligence Index leaderboard, Muse Wonder Full ranks just behind Claude Opus 5.5 Max, tying with Claude Sonnet 5.5 Max. This suggests that Meta has caught up to the AI frontier, previously lagging behind other tech giants.

Meta stated that Muse Wonder will soon power experiences in Muse, though it will be available only to paid subscribers. The new model family is designed for agentic AI and user safety, with Meta emphasizing that a full report on its safety evaluations will be released soon. Wang mentioned that the model family was developed under the codename "watermelon."

Muse Wonder comes in two versions: Muse Wonder and Muse Wonder Full, the latter being the more capable model. Active Focus and Active Memory are key features of Muse Wonder, enabling AI agents to maintain focus on complex tasks, recover from errors, and adapt to new information. Meta claims that Muse Wonder models will avoid sharing private information without permission and will refuse harmful requests while assisting with cybersecurity defense.

Written by urgent.news from Mashable's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Also reported by 6 other outlets

Read the original at mashable.com →

More in AI

More from Monday 12 October →