{
  "id": 14008663,
  "title": "Muse Wonder Puts Meta Back at the AI Frontier",
  "url": "https://urgent.news/2026/10/12/muse-wonder-puts-meta-back-at-the-ai-frontier",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-12T17:00:27.000Z",
  "source": {
    "name": "Wired",
    "slug": "wired",
    "url": "https://www.wired.com/story/muse-wonder-puts-meta-back-at-the-ai-frontier/"
  },
  "original_language": "en",
  "account": "Meta has unveiled its latest AI model, Muse Wonder Full, which boasts frontier capabilities and positions itself among the most capable AI models globally. According to Meta, the new models represent a significant leap forward on their scaling ladder, bringing them closer to personal superintelligence. These models will be accessible through an API and a tool named Muse Code, designed for software development using AI. The superior version of the model will power various features across Meta's platforms, including the Muse personal agent.\n\nMeta's pursuit of AI excellence has been marked by substantial investments and strategic moves. Last year, CEO Mark Zuckerberg reallocated billions to Meta Superintelligence Labs, hiring former Scale AI CEO Alexandr Wang to lead the initiative. Despite some research personnel leaving, Meta's efforts in AI have been relentless. The company previously led the AI landscape with its Llama series of open-weight models, which gained widespread adoption among researchers and startups. However, Meta has struggled to keep pace after releasing less impressive versions of the model, as rivals Anthropic, OpenAI, and Google introduced more advanced alternatives.\n\nIn April, Meta unveiled Muse Spark, its inaugural model from the newly formed Meta Superintelligence Labs. Additionally, the company released a compact open-weight model called Muse Glimmer and intends to roll out larger open-weight models in the future. The specifics of how Muse Wonder was constructed and trained have not been disclosed by Meta. Nonetheless, the company asserts that the largest iteration of Muse Wonder outperforms Anthropic's second-best model, Claude Fable 5.1, and is narrowly below its top model, Claude Opus 5.5, on the Artificial Analysis Intelligence Index—a widely recognized aggregate score of model performance. Meta further claims that Muse Wonder surpasses OpenAI's most powerful model, GPT-6 Astra, and outperforms both Claude Fable 5.1 and Claude Opus 5.5 on certain specialized benchmarks, including tasks involving legal documents and software creation. While benchmarks may not always accurately reflect real-world performance, this accomplishment marks a substantial advancement for Meta's models.\n\nMuse Wonder also incorporates two innovative features aimed at enhancing agent capabilities and reducing misbehavior. These features, named Active Memory and Active Focus, are designed to maintain the model's focus on a designated goal and prevent it from becoming confused. This is particularly crucial when AI undertakes complex research or large-scale software development, as such endeavors can lead to costly errors or compounding mistakes. Meta has invested considerable effort in training Muse Wonder to refuse generating hazardous outputs, such as instructions for creating bioweapons or guidance on hacking. The company demonstrates strong performance on benchmarks assessing model resilience to such prompts.\n\nRecent incidents involving rogue AI agents from Anthropic, OpenAI, and other organizations underscore the potential risks associated with powerful AI models when given open-ended commands. Meta has not disclosed whether Muse Wonder-powered agents have exhibited misbehavior during testing or the specific measures taken to mitigate this issue. Mark Zuckerberg announced on X that the company delayed the release of Muse models to focus on safety enhancements. Additionally, Meta unveiled a framework to ensure the safe scaling of model capabilities in a recent blog post. Muse Wonder's introduction may fuel an already intense competition among tech giants to develop increasingly capable AI systems, solidifying Meta's position as a formidable player in the rapidly evolving AI landscape.",
  "summary": "Meta’s biggest model rivals the world’s best—and includes new features to keep agents from going off the rails.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Wired Business",
        "title": "Muse Wonder Puts Meta Back at the AI Frontier",
        "url": "https://urgent.news/2026/10/12/muse-wonder-puts-meta-back-at-the-ai-frontier-14009371",
        "published": "2026-10-12T17:00:27.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}