Betting on Personal Agents to Surpass Rivals! Meta Launches New Model Muse Wonder, Zuckerberg Says It's at the Forefront of AI

Meta has launched the next generation artificial intelligence (AI) models, Muse Wonder and Muse Wonder Full, claiming they have reached industry-leading levels in both personal and professional agent tasks.

Meta CEO Mark Zuckerberg stated that this is an important step toward providing “personal superintelligence” for everyone. The new models will be used to upgrade Meta’s personal AI agent, Muse, but users need to subscribe to a paid service costing $20 or $100 per month to access Muse Wonder through Muse. Meta is trying to create an AI revenue stream outside of its advertising business with this move.

This release is a major achievement for Meta since it rebuilt its AI R&D system over the past year. In the past 14 months, Zuckerberg has significantly reorganized the AI team, recruited external talent, and continued to invest heavily in computing infrastructure. Meta AI lead Alexandr Wang said the new models are competitive with industry leaders and particularly excel in personal and professional agent tasks.

Compared to OpenAI and Anthropic’s recent focus on programming agents, Meta aims to differentiate by building personal agents that cover both daily life and professional work scenarios. However, performance advantages of these new models still need to be validated through external evaluations and actual user experience; whether they can persuade users to keep paying will also test Meta’s ability to turn AI investments into new revenue drivers.

Two new models unveiled, focused on personal and professional agents

On Monday, the 12th Eastern Time, Meta officially released the Muse Wonder series, including the high-performance Muse Wonder and the full-featured Muse Wonder Full. Zuckerberg stated on social media that Muse Wonder is larger in scale and more advanced in capabilities compared to the previous Muse Spark models, marking Meta’s entry to the forefront of AI capabilities.

Zuckerberg explained that since the establishment of Meta Superintelligence Labs over a year ago, Muse Spark was the first model series released by the team, while Muse Wonder represents a major upgrade. He called this release a critical step for Meta towards “putting personal superintelligence in everyone's hands.”

Meta hopes the new models will not only answer questions and generate content, but also better understand users' goals, remember important information, and continuously execute complex tasks. The company sees personal and professional agent work—covering programming as well as financial, legal, and other professional issues—as key use cases for Muse Wonder.

Currently, Meta provides developers with the new models via the Meta Model API and Muse Code. Muse Wonder will gradually be integrated into Meta’s personal AI agent Muse, with access first opening to subscription users; the company also plans to develop dedicated models to make these capabilities available to a broader user base in the future.

Subscription required for new models, Meta seeks AI revenue beyond advertising

According to reports, users need to subscribe to a paid service at either $20 or $100 per month to use Muse Wonder through the Muse AI agent app. This means that although Meta is seeking to bring AI agents to a wider consumer audience, access to the new models is not entirely free.

It’s important to note that the charges apply to the corresponding subscription services for using the new models in Muse, and do not mean there is no free version of the Muse app itself. Meta has already provided a free tier and two paid monthly plans at $20 and $100, giving users higher usage limits and more choices.

This pricing strategy indicates a potential change in Meta’s business model. For a long time, online advertising has been Meta’s core revenue source. Now, the company hopes to monetize AI agents capable of performing practical tasks through subscription and other new means.

Compared to traditional chatbots, the value of personal agents lies in completing tasks for users, not just providing information. For example, users can ask them to help plan trips, handle daily tasks, or complete multi-step workflows. If the models prove reliable enough, users may be more willing to pay for higher usage quotas and more complex task execution capabilities.

However, whether the subscription model will succeed still depends on whether users see enough value to justify the monthly $20 or even $100 fee. For Meta, launching new models is only the first step; user retention, conversion rates, and long-term engagement are the real tests for turning technological investment into revenue.

Meta claims parity with OpenAI, Anthropic; external validation remains key

According to reports, Meta claims Muse Wonder has the capacity to compete with OpenAI’s ChatGPT and Anthropic’s Claude on programming, finance, and legal question tasks.

Alexandr Wang said in an interview: “These are genuinely state-of-the-art models. We truly think they are competitive with the best models on the market, especially leading in personal and professional agent work.”

When asked about the industry’s doubt that Meta can truly return to the AI frontier, Wang expressed hope that users will try the products themselves and make their own judgments.

However, comparing model capabilities is not simple. Different models may excel at different tasks, and design and evaluation of benchmarks can affect results; test scores don’t always reflect real-world experience. Whether Muse Wonder continues to make progress in independent evaluations, developer adoption, and complex task performance remains to be seen.

This release is of particular importance for Meta. Some prior models did not gain broad acceptance in the developer community, and Muse Wonder is one of the key achievements since Zuckerberg reorganized the AI team. Only if the new models fulfill their performance promises in practice will Meta be able to turn its talent and infrastructure investments into sustainable technical competitiveness.

14-month AI system overhaul, Zuckerberg bets on superintelligence

This release comes after an unprecedented AI strategy readjustment by Meta over the past year. According to Bloomberg, Zuckerberg has refocused the company’s AI development through the last 14 months, invested massive funds in infrastructure for advanced model training and operation, and brought in Alexandr Wang through an approximately $14 billion investment in Scale AI, tasking him with leading the new AI team.

Meta hopes this round of rebuilding will re-establish its competitiveness in frontier AI. Rather than simply chasing larger parameter models or higher benchmark scores, the company is also trying to integrate advanced models with its massive consumer product ecosystem, bringing AI capabilities into daily usage scenarios.

Competition in generative AI is shifting from chatbots that simply answer questions or generate content to agents capable of invoking tools and continuously executing tasks. The latter require models with stronger planning, context management, and long-term task execution abilities, and are also expected to become the main touchpoint for consumers to interact with AI services.

Meta’s advantage is its close integration with users’ daily lives through Facebook, Instagram, WhatsApp and other platforms. If it can successfully integrate advanced models into these products and enable AI agents to truly complete tasks for users, Meta will have the opportunity to broaden its user base and explore new monetization paths such as subscriptions.

Different from focusing solely on programming agents, Meta aims to build personal AI assistants

Alexandr Wang stated in interviews that while other AI industry players now focus heavily on programming agents, Meta wants to chart a different strategy, unlocking the practical value of personal agent applications.

This approach is already reflected in Meta’s recent product portfolio. The company released the Muse personal AI agent with the aim of helping users handle daily tasks and pursue personal goals. Muse Wonder will empower this product with stronger model capabilities to tackle more complex tasks.

However, successfully scaling personal agents requires more than strong model capabilities—it also needs solutions for reliability, privacy, and permission management. When users ask AI to manage emails, calendars, or other personal affairs, the models must not only provide correct answers, but also reliably execute actions within appropriate authorization.

For Meta, whether Muse Wonder can help the company reclaim a leadership position in AI ultimately depends on if its performance converts to genuine developer demand and user adoption; and whether the $20 and $100 per month subscription plans earn market recognition will further test the company’s AI monetization strategy.