Meta Launches Muse Wonder AI Models for Personal and Professional Tasks

What Meta Announced

Meta said on October 12, 2026 that it had released Muse Wonder and Muse Wonder Full, calling them "our most capable models to date, built to deliver state-of-the-art performance on personal and professional agentic tasks," according to the company's own announcement on its Meta Newsroom blog.

The models are designed to handle multi-step tasks rather than single replies. Meta's announcement lists navigating complex documents such as insurance claims, leases and financial records, reasoning in expert domains including law, finance and health, and coding for apps and games as core use cases.

Two features underpin the release: Active Memory, which keeps track of relevant details across a conversation, and Active Focus, which lets the model keep working toward a goal, recover from errors and ask clarifying questions when it gets stuck, Meta said.

Demos and Benchmarks

On its research blog, Meta showed four example tasks: building a playable 3D kitchen game from uploaded images, assembling a personalized running plan using map data, designing and ordering custom merchandise from photos taken from an Instagram profile, and producing a wedding seating chart based on a list of matching requirements.

Independent evaluator Artificial Analysis scored Wonder Full at 56 and Wonder at 52 on its Intelligence Index, according to VentureBeat. The report said Wonder Full tied Anthropic's Claude Sonnet 5.5 and scored above OpenAI's GPT-6 Astra on that index, while Claude Opus 5.5 ranked highest among the models compared, at 58. On the SWE-Marathon coding benchmark, VentureBeat reported Wonder Full scored 57.5 versus 46.3 for Opus 5.5. Meta's own account, reported by TechCrunch, described the two models as achieving benchmark scores comparable to Fable 5.1 and GPT-6 Astra.

Pricing and Access

Meta is selling access to the smaller Wonder model at $4 per million input tokens and $20 per million output tokens through its API, according to TechCrunch and VentureBeat. TechCrunch said that price is roughly equivalent to OpenAI's GPT-6 Sol or Anthropic's Claude Opus 5.5, while VentureBeat reported the same rate matches Claude Opus 5.5 but undercuts OpenAI's GPT-6 Astra, which VentureBeat listed at $10 per million input tokens and $50 per million output tokens. Wonder Full is reserved for subscribers of Muse Code, Meta's coding-assistant subscription, and is not sold through the public API, according to VentureBeat.

Competitive Landscape and Market Reaction

TechCrunch described the release as Meta's first frontier-class model launch since August 2025, a period in which it said Meta had avoided the kind of state-of-the-art model releases OpenAI and Anthropic are known for. Both models were trained at Meta's Prometheus data center, which Meta's own research blog described as a gigawatt-scale facility; TechCrunch reported the center is located in central Ohio and came online earlier in 2026. TechCrunch noted Wonder is positioned as a larger, more capable successor to Meta's earlier, lighter-weight Muse Spark model.

GuruFocus reported that the Muse app reached the top of Apple's App Store charts after the launch, and noted Meta's platforms reach roughly 3.6 billion daily active users across Facebook, Instagram, WhatsApp, Messenger and Threads. GuruFocus cited analyst Dan Ives as saying the new capabilities could make Meta's assistant a more effective problem-solving tool and help accelerate user adoption. The same GuruFocus report noted Meta shares traded at $759.35, which GuruFocus calculated as about 12.3% below its estimated intrinsic value of $865.38, and said company insiders had sold $260.3 million in shares over the prior 12 months with no reported insider purchases in that period.