Meta Releases AI Model Muse Wonder as Zuckerberg Says Benchmarks Target OpenAI and Claude

TradingKey - Meta Platforms (META) released the AI model Muse Wonder on Monday, claiming it possesses "frontier capabilities." This release will test the technical results of Meta's multi-billion-dollar AI reorganization and reflects the company's efforts to generate AI revenue beyond its core advertising business.

It is reported that Muse Wonder, internally codenamed "Watermelon," is the company's most capable model since Zuckerberg aggressively recruited talent and reorganized its AI business. Meta stated that the model rivals OpenAI's ChatGPT and Anthropic's Claude in handling coding, finance, and legal tasks, while outperforming Claude Opus 5.5, Fable 5.1, and GPT-6 Astra in certain benchmarks.

As of press time, Meta shares were up 4.13% at $748.34.

META stock price chart, Source: TradingView

Alexandr Wang, head of Meta AI, stated that the new model is particularly adept at "personal and professional agent" tasks. The team trained the model for scenarios such as making restaurant reservations and checking availability, enabling it to "try alternative methods" when encountering obstacles, avoiding getting stuck and repeatedly questioning users. He also urged skeptics to "try it for themselves and form their own judgment."

Fees cover personal applications and developer products. The Muse app itself is free, with subscriptions offering additional usage quotas; the more powerful Muse Wonder Full is available exclusively to Muse Code developer subscribers, priced at $50 or $200 per month. Developers can also pay to access Muse Wonder via the Meta Model API. Wang noted that while API access costs more than standard subscriptions, pricing will remain "significantly lower" than leading models from OpenAI and Anthropic.

The enterprise market is also a target for Meta. Zuckerberg stated that reaching the AI frontier requires vast computing power, infrastructure, and top talent, and the company will continue to scale up investment and advance "personal superintelligence." Unlike the path pursued by OpenAI and Anthropic toward artificial general intelligence, Meta aims to build assistants that empower individuals rather than systems designed primarily to automate knowledge work.

Regarding safety concerns, Zuckerberg opposed a broad industry slowdown of AI development, arguing that individual labs should ensure the safety of their models. Meta stated that the new model offers the company's "strongest-to-date" protections against online abuse, which are currently available only with paid models.