{
  "id": 14017946,
  "title": "Meta releases its first frontier model, Muse Wonder",
  "url": "https://urgent.news/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-12T17:56:56.000Z",
  "source": {
    "name": "TechCrunch",
    "slug": "techcrunch",
    "url": "https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"
  },
  "original_language": "en",
  "account": "Meta's Superintelligence Lab has unveiled its first frontier model, Muse Wonder, alongside Muse Wonder Full, marking the company's first foray into creating highly capable, large-scale models. While Muse Spark had been designed to be lightweight and fast, the Wonder models are significantly larger, boasting higher benchmark scores and superior performance on complex tasks such as coding. According to internal testing, Wonder and Wonder Full match the benchmark scores of Fable 5.1 and GPT-6 Astra. The release represents a significant scaling milestone for Meta, bringing it closer to achieving personal superintelligence. Pricing for the models through the Meta Model API is $4.00 per million input tokens and $20.00 per million output tokens, positioning them similarly to GPT-6 Sol or Opus 5.5. However, Wonder Full is exclusive to Muse Code subscribers and will not be accessible via the API. Both models will also power the Muse agent, with Meta focusing on enhancing the models' \"active memory\" capabilities, or their ability to retain and synthesize user-provided information over extended periods. This focus on active memory is demonstrated by the MemoryBench benchmark, where Wonder now outperforms Fable, Opus, or Astra. Research scientist Kavya Kopparapu highlighted the company's emphasis on creating models that can remember important details over long periods, from days to years. The new frontier models are the first of their kind since Meta's AI efforts were restructured by former Scale AI founder Alexandr Wang in August of last year. Since then, Meta has introduced the lightweight Muse Spark model and the Muse agent, but has not made widely publicized, state-of-the-art model releases as frequently as OpenAI or Anthropic. The models were trained at Meta's Prometheus data center in central Ohio, which began operations earlier this year.",
  "summary": "Meta's Superintelligence Lab released a pair of new state-of-the-art models on Monday, Muse Wonder and Muse Wonder Full, the company's first attempt at a frontier model.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Techmeme",
        "title": "Muse Wonder Full (max) outperforms Fable 5.1 (max) on Artificial Analysis' Intelligence Index, matches Sonnet 5.5 (max), and is just shy of Opus 5.5 (max) (@artificialanlys)",
        "url": "https://urgent.news/2026/10/12/muse-wonder-full-max-outperforms-fable-5-1-max-on-artificial-analysis",
        "published": "2026-10-12T17:50:01.000Z"
      }
    ]
  },
  "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."
}