Zuck's Muse Wonder Drops: Meta Joins the AI Arms Race Nobody Asked For 🏛️

Meta released two new artificial intelligence models, Muse Wonder and Muse Wonder Full, on Monday, calling them its most capable models to date and positioning them as direct competitors to flagship offerings from OpenAI and Anthropic. The launch marks the second model family produced by Meta Superintelligence Labs, following the April debut of Muse Spark, and signals Meta's escalation of capital and engineering toward so-called "personal superintelligence."

"Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence," Meta wrote in a blog post. "Muse Wonder is a major leap towards delivering personal superintelligence for everyone," CEO Mark Zuckerberg wrote on Threads. "Congrats to everyone on our team for achieving this milestone. I'm looking forward to seeing how you all use it." The company's use of the term tracks President Trump's stated view, expressed on Truth Social last Thursday, that anyone using "Artificial Intelligence" instead of "Super Intelligence" should be considered "the enemy." Meta has called its AI division Meta Superintelligence Labs since 2025.

Both models carry a 1 million token context window, roughly 750,000 words, per Meta's model card. Muse Wonder is available now through the Meta Model API and Muse Code, Meta's coding agent, while Muse Wonder Full is restricted to paying Muse Code subscribers. Two features anchor the release: Active Memory, which carries relevant information across separate conversations, and Active Focus, which sustains work on difficult tasks by trying new approaches and checking with the user before major steps. The standard Muse Wonder is priced at $4 per million input tokens, matching Claude Opus 5.5 and undercutting GPT-6 Astra at $10 and Claude Fable 5.1 at $50, with Wonder Full available only to subscribers on Muse Code's $50 monthly "power usage" or $200 monthly "maximum usage" tiers.

Benchmark claims diverged across evaluators. On Meta's internal Personal Superintelligence Index, an in-house composite of personal and professional assistant tasks whose exact formula has not been published, Wonder Full scored 59.1, ahead of Claude Opus 5.5 at 55.2, GPT-6 Astra at 54.5 and Claude Fable 5.1 at 49.9. On the SWE-Marathon coding benchmark, Wonder Full scored 57.5 versus 46.3 for Opus 5.5 and 42.5 for Astra. Independent platform Artificial Analysis scored Wonder Full at 56 and the standard Wonder at 52 at maximum reasoning effort, placing Full above Claude Fable 5.1 and GPT-6 Astra, which each scored 53, but below Claude Opus 5.5's 58. Abacus.AI CEO Bindu Reddy pushed back on the claims, writing on X that Muse Wonder was "nowhere near Fable or Astra level" and promising detailed benchmark results.

The release comes weeks after OpenAI shipped GPT-6 Astra on September 3 and Anthropic rolled out Claude Opus 5.5 on September 22, the latter matched by Meta as the competitive bar. It also follows Meta's roughly $14 billion deal for Scale AI, whose founder Alexandr Wang was placed in charge of Meta Superintelligence Labs, and the death of the open-source Llama family after Muse Spark's closed, smaller launch in April. Musk's SpaceXAI released Grok 4.7 last month to match the latest rival models, as the frontier AI race continues to absorb capital across the industry.