Meta has unveiled two new artificial intelligence models, Muse Wonder and Muse Wonder Full, which the company called its most capable models to date and a step toward what co-founder Mark Zuckerberg has described as “personal superintelligence.” The announcement was published Monday on Meta's research blog, with Meta claiming the new systems rival leading offerings from OpenAI and Anthropic. The release puts Meta's in-house models in direct competition with the industry's leading AI labs, making independent benchmarks and pricing the early yardsticks for whether the company's framing holds up.

The models introduce two capabilities that Meta calls Active Memory and Active Focus. Active Memory carries relevant information across conversations, allowing context to persist between exchanges, while Active Focus is designed to help the agent sustain work on difficult tasks over time. Both capabilities point toward agentic AI — systems built to carry out extended, multi-step work rather than one-off exchanges — a focus that has become increasingly prominent across the AI industry.

Independent evaluation platform Artificial Analysis, which published its own assessment of the release, scored Wonder Full at 56 on its Intelligence Index and the standard Wonder model at 52, with both results measured at maximum reasoning effort. Those scores place Wonder Full above Claude Fable 5.1 and GPT-6 Astra, which each scored 53, but below Claude Opus 5.5, which scored 58. The standard Wonder model, at 52, fell below all three of those rival systems. Third-party rankings like these have become a widely cited reference point for weighing vendors' launch claims against measured performance.

The performance claims drew pushback. Bindu Reddy, CEO of Abacus.AI, which conducts its own model evaluations, challenged Meta's characterization in a post on X, writing that Muse Wonder was “nowhere near Fable or Astra level” and saying that detailed benchmark results would follow. Those promised results would add another independent data point for readers weighing Meta's claims against Artificial Analysis's scores.

Pricing makes the new models Meta's most expensive to date. The standard Wonder model charges $4 per million input tokens, matching Claude Opus 5.5 while undercutting GPT-6 Astra and Claude Fable 5.1, which charge $10 and $50 per million input tokens, respectively. Per-million-token input pricing is the standard unit for comparing commercial model costs, making those figures a like-for-like measure of where Wonder sits on cost. Access to Wonder Full is limited to subscribers of Muse Code's $50-per-month “power usage” tier or its $200-per-month “maximum usage” tier.

This article was originally reported by Cointelegraph on Oct. 13, 2026, under the headline “Meta Launches Muse Wonder AI Models Competing with Rivals.”