With the announcement of Muse Spark 1.1, Meta has advanced its ambitions in artificial intelligence. Released on Thursday, the multimodal reasoning model is intended to provide an advance in reasoning, automation, and tool usage in both consumer and developer products.
However, the importance of the product goes beyond technical aspects. Meta has made it clear that the model would be the first of its AI solutions to have a paid tier for developers. This marks a major step toward generating revenue from AI solutions.
What Muse Spark 1.1 Actually Does
According to Meta, the multimodal reasoning model has drastically improved compared to previous ones in regard to tool usage, computer usage, coding, and multimodal capabilities. The company claims that the model is able to maintain context across sessions and adapt to changing requirements during the completion of a task.
Meta states that Muse Spark 1.1 has the ability to navigate through an unfamiliar software interface with minimal human assistance, something that the company believes is crucial in terms of the agentic artificial intelligence it has developed.
As stated by the company, Muse Spark 1.1 was trained to assess when to create a script rather than complete the actions in question by clicking. The model was created to understand “when to automate and when to use the interface directly,” finding the shortest way instead of completing the entire chain of actions step by step.
Moreover, the company has presented Muse Spark 1.1 along with its Muse Image tool released earlier in the week as an element of a broader vision of creating personal superintelligence. In particular, Meta wants to develop AI tools that would assist people in pursuing their goals, building their visions, and strengthening relationships as well as taking action on behalf of people.
In benchmarking tests, Meta claimed that Muse Spark 1.1 demonstrated better results than several competitors’ AI models in coding and agentic response times. However, no detailed figures were released together with the announcement of the product.
Why Agentic AI Raises Concerns
The agentic capabilities of Muse Spark 1.1 are noteworthy as the company intends to develop AI tools that are able to act on behalf of a user through various apps and platforms. According to Vishal Shah, the vice president of AI products at Meta, Muse Spark 1.1 should guarantee high-quality performance of personal tasks that involve planning and coordinating activities across multiple external platforms.
The issue with such capabilities is security, given the multiple examples of flawed AI decision-making. For instance, it was recently revealed that Meta’s agentic AI support agents on Instagram accidentally granted hackers access to about 20,000 accounts after being asked to do so. Such an incident is one of many that showed a negative outcome of flawed AI decision-making.
However, despite its history of problems with the agentic AI, Meta is confident that Muse Spark 1.1 won’t repeat the mistakes of its predecessor.
The Money Question Behind the Launch
The existence of the paid tier for Muse Spark 1.1 implies the importance of generating revenue from the technology. As reported by Bloomberg, CEO of Meta, Mark Zuckerberg, claims that pricing of the model would be attractive and aggressive to attract developers from other AI platforms as soon as possible.
Such rush can be explained by the scale of Meta’s expenses related to artificial intelligence. The company has already invested hundreds of billions of dollars into AI infrastructure and development, and it will likely invest more than a trillion dollars in total. According to the company, such an expense is a transformative bet on the future of its business.
It goes without saying that the math behind such a bet is quite hard. Even if it generated 100 billion dollars a year from AI subscriptions, it would take more than 10 years to earn back all expenses related to AI solutions, without accounting for further maintenance of these systems. Meta has to develop multiple AI products to justify its investment to consumers and enterprises.
To provide some context, Meta earned about 200 billion dollars in 2025 with nearly 98 percent of the earnings from advertising. For Meta, this means it has to create a whole new revenue source from scratch, one that does not include advertising, to make its AI bet profitable.
A Bigger Bet on Business Automation
Meta has to focus on monetization strategies such as subscriptions and business access to its top AI solutions to persuade enterprises that the investment is worthwhile. To accomplish that, the company needs its agentic AI solutions to effectively substitute human workers, even in cases where the technology is not fully ready for it.
The intention to do so has been already revealed by Meta. In particular, according to the plans of the company, it expects to replace 90 percent of its content review workers with AI by the end of 2026. However, recent reports claim that the current level of effectiveness of AI tools is much lower than expected by companies.
As more businesses implement agentic AI systems while human employees remain responsible for their implementation and functioning, the gap between expectations and performance is likely to lead to increased workplace friction, regardless of how the system actually performs.
Muse Spark 1.1 is Meta’s most significant bet on closing this gap. The success of the product will not only define the direction of Meta’s AI development but will also show how much confidence enterprises place in autonomous AI solutions.
How Muse Spark 1.1 Fits Into the Broader AI Race
Meta is not the only company racing to build powerful agentic AI capable of planning, reasoning, and action without human intervention. Other competing labs have been developing similar tools meant to automate various processes, including customer support and software development. Meta’s decision to introduce a paid tier to Muse Spark 1.1 is likely aimed at positioning itself as a serious commercial player in this area rather than an experimental research lab trying to explore new opportunities.
This move matters because Meta has traditionally provided most of its AI work through open access, as seen with its earlier Llama models. The introduction of a paid developer tier signals a shift to the typical software business model where advanced features are accessible only to paying customers while lighter versions stay available for free or open access. This move also creates an expectation that Muse Spark 1.1 should provide enough value to be paid for since most businesses would be reluctant to delegate operational control to an AI system at this point.
What Comes Next for Meta’s AI Strategy
At this point, it is uncertain whether Muse Spark 1.1 will be followed by any significant updates soon; however, the number of new releases this year suggests that Meta is not planning on slowing down. Between releasing Muse Image and Muse Spark 1.1 within a week of each other, Meta seems to consider fast iteration a competitive edge, and the company keeps providing new features even when there are still doubts about their safety and reliability.
As far as the immediate future goes, the most important factor is financial because Meta needs to see the ROI of its AI investments. With the paid tier being introduced for Muse Spark 1.1, it becomes clear how the test results of this tool could affect the future terms of Meta’s AI releases.
Frequently Asked Questions
Q 1: What is Muse Spark 1.1?
Muse Spark 1.1 is Meta’s new multimodal reasoning model designed for agentic tasks, including improved tool use, coding, and computer automation.
Q 2: Does Muse Spark 1.1 cost money to use?
Yes. Meta confirmed it will be the first of its AI models to include a paid tier specifically for developers.
Q 3: What can Muse Spark 1.1 do differently from earlier models?
It can maintain context across sessions, adapt to changing requirements, and navigate unfamiliar software interfaces with minimal human input.
Q 4: How does Muse Spark 1.1 relate to Muse Image?
Meta launched both models within the same week as part of its broader vision for personal superintelligence tools.
Q 5: What risks come with agentic AI like Muse Spark 1.1?
Agentic AI models can make mistakes while acting independently, as shown when Meta’s AI support agents mistakenly gave hackers access to thousands of Instagram accounts.
Q 6: Why is Meta introducing a paid tier now?
Meta is under pressure to generate revenue from its massive AI spending, and a paid developer tier is one of several planned monetization paths.
Q 7: How much has Meta spent on AI so far?
Meta has committed hundreds of billions of dollars, with some estimates suggesting spending could eventually surpass a trillion dollars.
Q 8: Is AI actually saving businesses money right now?
Recent reports suggest AI tools are not yet delivering the major efficiency gains companies were promised, raising questions about near-term returns.
Q 9: What is Meta’s plan for content moderation staff?
Meta has said it aims to replace 90 percent of its content review staff with AI tools by the end of 2026.
Q 10: What does Meta’s revenue currently depend on?
Meta reported about 200 billion dollars in revenue for 2025, with roughly 98 percent coming from advertising.
