Meta spent the better part of two years talking about a future where every person on Earth has a capable AI agent working on their behalf. On September 8, 2026, that future got a name, a price tag, and a payment-card requirement. Meta introduced Muse, calling it “a secure, private personal AI agent” and, more boldly, “the world’s first personal AI agent built for everyone,” according to the company’s official announcement, reported by outlets including an AP dispatch carried by VIN News. The announcement lands at a moment when OpenAI, Google, Anthropic, and Amazon are all racing to define what a personal AI agent actually does, and Meta just bet its distribution advantage, 3 billion-plus app users, on being first to make the category concrete.

What Meta Actually Announced on September 8Pricing: Free Tier, Power at $20, Maximum at $100Muse at a GlanceThe Security Pitch: Isolated Machines and Opt-Out TrainingMuse Spark: The Model Family Behind the AgentMeta Superintelligence Labs and the Push for a Frontier ModelHow Muse Compares to the Rest of the Agent MarketWall Street’s Read on the LaunchFrom Meta AI to Muse: The Longer ArcWhat This Means for the AI Agent MarketOpen Questions and RisksPredictions: Where Muse Goes From HereThe Bottom LineFrequently Asked QuestionsWhat is Meta Muse?How much does Meta Muse cost?Is Meta Muse available outside the United States?Do I need a credit card to use the free tier?What model powers Meta Muse?Can I stop Muse from using my data to train Meta’s AI models?How does Muse compare to ChatGPT Atlas or Grok 4.5?Is there advertising inside Muse?Related Coverage

The pitch is simple on its face. Muse is supposed to help people “stay on top of things, takes tasks and projects off their plate, and turns long-term goals into action plans,” according to the official announcement. What follows is less simple: a tiered pricing structure, a weekly token allowance tied directly to Mark Zuckerberg’s own words, a mandatory payment card at signup, and a security model built around isolated cloud machines. This piece breaks down what actually shipped, how it stacks up against rival agents, and what the launch signals about where the AI agent market is headed next.

The core announcement is Muse, positioned not as a chatbot but as an agent that acts. Meta’s own product page, How We Designed Muse, published the same day on introducing.muse.ai and referenced on the company’s broader Meta AI blog, frames the tool around task completion rather than conversation. Reports describe Muse handling everyday chores such as drafting and sending emails, booking travel, buying tickets, filling out forms, negotiating bills, and building longer-term plans that stretch across weeks or months rather than a single session. Meta said as much in its own words: “Introducing Muse, your personal AI agent from Meta that gets things done across every part of life.”

Muse ships across four surfaces: the web, iOS, Android, and WhatsApp. That last channel matters more than it looks. WhatsApp alone counts billions of monthly users, and folding an agent directly into a messaging app that people already open dozens of times a day gives Meta a distribution channel no pure-play AI lab can match. Reports note that support for Meta’s AI glasses hardware is coming later rather than at launch, which suggests the company is sequencing its rollout rather than shipping everything at once.

At launch, Muse is available in the United States only, and multiple reports describe an adults-only restriction, tightening the rollout further. That’s a narrower opening than Meta AI’s original rollout across its family of apps, and it points to a company being more careful this time around, likely a response to the regulatory heat Meta has taken over teen safety and data handling in prior product launches, a topic covered in our look at how ChatGPT’s teen safety rules compare to Meta’s own record.

Pricing: Free Tier, Power at $20, Maximum at $100

Meta is running Muse on a freemium ladder. A free tier sits underneath two paid plans, reportedly named Power at $20 a month and Maximum at $100 a month. That structure mirrors what OpenAI and Anthropic have already normalized in the market, a low-cost consumer tier alongside a premium tier aimed at power users who burn through far more compute.

The free tier isn’t unlimited, but it’s generous. Zuckerberg said Muse will be usable for free “up to 100 million tokens per week” per user, a figure that puts Meta’s free allowance well above what most rivals disclose publicly for their no-cost tiers. Whether that number holds steady once usage scales past early adopters is an open question, and it’s the kind of promise companies have trimmed before once real-world demand hits.

One detail stands out as unusual for a “free” product: reports say Muse requires a payment card on file just to get started, even on the no-cost tier. That’s a meaningful friction point. It suggests Meta wants to curb bot signups and abuse from day one, but it also means the free tier isn’t the zero-commitment trial that “free” usually implies elsewhere in consumer software.

Muse at a Glance

DetailWhat’s reported
Announcement dateSeptember 8, 2026
Launch marketUnited States only
Age requirementAdults, 18 and older
Access channelsWeb, iOS, Android, WhatsApp
Free tier limitUp to 100 million tokens per week (per Zuckerberg)
Power plan$20 per month
Maximum plan$100 per month
Signup requirementPayment card required, even for the free tier
Underlying model familyMuse Spark
Security architectureDedicated virtual machine per user in Meta’s cloud
Ads at launchNone reported, though Meta is said to be exploring commerce features
Hardware integrationAI glasses support planned for a later date

The Security Pitch: Isolated Machines and Opt-Out Training

Meta is leaning hard on privacy and security language for this launch, more than it has for any consumer AI product to date. Reports describe Muse running inside a dedicated virtual machine in Meta’s cloud, effectively isolating each user’s agent session and data rather than pooling everything into a shared environment. Zuckerberg put it directly in the announcement: “You choose which apps and services Muse has access to and you can disconnect them at any time.”

Users reportedly also get the option to opt out of having their Muse interactions used to train Meta’s future AI models, a control that wasn’t always front and center in earlier Meta AI rollouts. Meta says it plans to ship an encrypted version of Muse later in 2026, which would extend the privacy model further by locking even Meta itself out of session content by default.

That security framing isn’t just marketing polish. An agent that can send emails, buy tickets, and negotiate bills needs access to a person’s inbox, calendar, and payment details to function at all. Giving an AI system that kind of reach is a much bigger trust ask than a chatbot that only answers questions, and Meta clearly knows it. The company’s emphasis on isolation and opt-outs reads as a direct answer to the kind of skepticism that’s dogged agentic AI rollouts elsewhere in the industry, including the caution other labs have shown around opaque reasoning in agentic systems.

Muse Spark: The Model Family Behind the Agent

Muse runs on Meta’s Muse Spark model family, the same lineage that Meta Superintelligence Labs has been quietly building out for months. Our earlier coverage tracked the rollout of Muse Spark 1.3, which cut tool-call counts by roughly 20% and token usage by about 25% in agentic coding tasks, and Meta said at the time that “Muse Spark 1.3 will roll out Wednesday,” a rare instance of the company giving a hard ship date ahead of a model release.

That efficiency push matters directly for the consumer product. An agent that burns fewer tokens per completed task stretches a free tier’s 100-million-token weekly budget further, and it lowers Meta’s own compute bill at a moment when GPU and memory costs are climbing across the industry. Meta has also been shipping smaller, faster capabilities under the same umbrella, including the low-latency Muse Voice Transcribe engine built for AI glasses, which points to a broader strategy of building Muse-branded components that eventually converge into one agent experience across text, voice, and hardware.

Meta Superintelligence Labs and the Push for a Frontier Model

Muse’s model stack sits under Meta Superintelligence Labs, the newer research and product group led by Chief AI Officer Alexandr Wang, who joined Meta after the company’s investment in his previous company, Scale AI. Reporting through the spring and summer of 2026 described Muse Spark as Meta’s first real attempt at a frontier-class model family under that structure, aimed at closing the gap with OpenAI and Anthropic rather than simply keeping pace through open releases like Llama.

That’s a notable shift in posture. Llama built Meta’s reputation as the industry’s open-model champion, giving developers weights they could run and fine-tune themselves. Muse flips that script. It’s a closed, subscription-priced consumer product, run entirely on Meta’s own infrastructure, with no equivalent open release attached. The company appears to be running two AI strategies in parallel now, open models for developers and researchers, and closed agents for consumers willing to pay.

How Muse Compares to the Rest of the Agent Market

Muse doesn’t launch into an empty field. OpenAI, xAI, Google, and Anthropic have all shipped their own agentic products in the past several months, and each one is chasing a slightly different version of the same idea: an AI system that doesn’t just answer questions but completes tasks. The table below lines up what’s publicly known about pricing and positioning as of this month.

ProductCompanyPricingAccessPositioning
MuseMetaFree (100M tokens/week) + $20 Power / $100 MaximumWeb, iOS, Android, WhatsAppConsumer life-management agent tied to Meta’s social apps
GPT-6 AstraOpenAI$10 / $50 tiersWeb, apps, APIGeneral-purpose agent with broad task and coding reach
Grok 4.5xAI$2 / $6 tiersX app, webLower-cost agent aimed at coding and real-time tasks
Claude (agentic tools)AnthropicNot directly comparableWeb, apps, APIPositioned around knowledge work, coding, and enterprise use
Alexa+AmazonNot directly comparableEcho devices, appVoice-first, home and household focus

The gap between $2 and $100 a month across these products shows how unsettled this market still is. Nobody has agreed on what a personal agent should cost, because nobody has agreed on what one is actually worth to an average user yet. Grok 4.5’s entry-level pricing undercuts Muse’s Power tier by a wide margin, while GPT-6 Astra sits closer to it. Meta’s bet is that its $100 Maximum tier, paired with deep access to a user’s calendar, inbox, and messaging habits through Facebook, Instagram, and WhatsApp, justifies a premium that standalone agent products can’t match.

Anthropic and Amazon occupy different lanes entirely. Claude’s agentic tools skew toward developers and enterprise workflows rather than consumer errand-running, and Alexa+ still leans on voice and smart-home integration as its core differentiator. Muse is explicitly chasing neither of those lanes. It wants to be the thing that books your flight, drafts the email to your landlord, and tracks a goal you set three weeks ago, all from inside an app you already have open. Ongoing coverage of Meta’s product strategy, including The Verge’s reporting on Meta, has tracked this shift toward deeper app-level integration over the past year.

Wall Street’s Read on the Launch

The market reaction leaned positive in the window around the announcement, with Meta shares climbing roughly 3% to 4%, according to multiple market reports. Bank of America reportedly maintained a Buy rating and pointed to meaningful upside in a note issued around the launch, framing Muse as early evidence that Meta’s enormous AI infrastructure spending is starting to translate into shippable products rather than just capacity. Morgan Stanley went further, reportedly describing the launch as the first step in re-rating the stock, a signal that at least some analysts see Muse as a turning point rather than a routine product drop.

Not every note was uniformly bullish. Bernstein and KeyBanc were described as constructive on the launch, though KeyBanc reportedly trimmed its price target even while staying positive overall, a sign that some caution remains about how quickly Muse can turn into meaningful revenue. Reuters-based reporting also surfaced mixed feedback from internal testing, including praise for the tool’s travel-planning usefulness alongside concerns about dropped sessions and the risk of exposing sensitive data, the kind of rough edges that tend to show up whenever a company moves an AI product from lab testing to a live consumer launch.

Alexandr Wang addressed the pricing structure directly in comments reported by CNBC. “The Muse personal agent will be available in either a free tier or through monthly subscription plans of $20 or $100, depending on usage,” Wang said, according to CNBC’s coverage of the launch. That framing puts usage, not features, at the center of Meta’s pricing logic, a departure from the flat feature-gated tiers most software companies still default to.

Muse doesn’t come out of nowhere. Meta AI has served as the company’s consumer-facing AI layer across Facebook, Instagram, and WhatsApp for a few years now, and Meta has repeatedly pointed to that assistant, plus its smart glasses line and ad-targeting tools, as key pieces of its future cash flow story. Llama, meanwhile, built Meta’s credibility in the open-source AI world, giving the company a seat at the table even during stretches when its consumer AI products lagged the competition on raw capability.

What changed is the emergence of Meta Superintelligence Labs as a distinct unit chasing frontier-level performance rather than just open, freely available weights. Muse Spark, shipped under that unit, was described by analysts earlier this year as a serious attempt to reduce uncertainty about whether Meta could genuinely compete with OpenAI and Anthropic on model quality, not just on distribution. Muse itself is the natural next step in that arc: a consumer product built on top of a model stack that Meta now believes is strong enough to stand behind at scale, wrapped in the kind of security and privacy language the company knows it needs after years of data-handling controversies.

Meta’s own announcement framed the ambition plainly. As Zuckerberg put it in comments widely quoted across launch-day coverage, “Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about. Your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more.” Meta’s corporate newsroom, About Meta, has published similar framing around the company’s broader personal-superintelligence push over the past year. It’s a sweeping claim, and one Meta will have to back up with a product that actually holds up once millions of people, not just internal testers, start trusting it with their inboxes and calendars.

What This Means for the AI Agent Market

The immediate market impact is competitive pressure. Meta just validated the idea that consumers will pay real money, up to $100 a month, for an agent that manages parts of their life rather than just chatting with them. That’s a useful data point for every other lab watching from the sidelines, and it raises the stakes for OpenAI, Google, and Anthropic to show their own agent products can hold onto users at similar or higher price points.

It also puts fresh pressure on infrastructure. Personal agents that run persistently, book things, send messages, and check in on long-term goals need far more sustained compute than a one-off chatbot reply. That demand lands right on top of an industry already dealing with tight GPU and memory supply, a dynamic our coverage has tracked closely as prices for high-end accelerators and server memory have climbed through 2026. Meta’s willingness to charge $100 a month for its top tier suggests the company is pricing in real infrastructure cost, not just testing what the market will bear.

There’s a distribution angle too, and it might be the most important one long-term. Meta doesn’t need to win users over from scratch. It can slot Muse directly into WhatsApp conversations billions of people already have every day, a channel none of its rivals can replicate at the same scale. That single advantage could matter more than any benchmark score once the agent race moves from “who’s smartest” to “who actually gets used.”

Open Questions and Risks

A few things remain genuinely unresolved. Meta hasn’t published independent benchmark comparisons showing how Muse Spark stacks up against GPT-6 Astra, Grok 4.5, or Claude on standardized agentic tasks, so claims about its capability level are, for now, coming largely from Meta itself. The mandatory payment card on the free tier is also a meaningful trust hurdle that Meta will need to justify clearly, since it blurs the line between “free” and “trial.”

The Reuters-reported internal testing concerns about dropped sessions and possible sensitive-data exposure are worth watching closely too. An agent that handles bill negotiation and form-filling is, by definition, touching financial and personal data constantly, and any early stumbles there could do lasting damage to user trust well before the product has a chance to mature. Meta’s promised encrypted mode, expected later in 2026, will be one signal of how seriously the company is treating that risk.

Predictions: Where Muse Goes From Here

Expansion beyond the US within two quarters. Meta rarely keeps a major consumer AI product US-only for long once early metrics come in, and the WhatsApp channel alone gives it strong reason to push into international markets fast.
AI glasses integration becomes the next big Muse milestone. Meta has already built the low-latency voice pipeline for its glasses hardware, and folding full Muse agent access into that hardware would be a differentiator none of its rivals can match without their own consumer hardware line.
Pricing gets adjusted within 90 days. With Grok 4.5 undercutting Muse’s Power tier at $2 to $6 a month and GPT-6 Astra sitting closer at $10 to $50, expect Meta to tweak either the free-tier token cap or the Power tier price once real usage data comes in.
Encrypted mode ships as a trust-rebuilding move. Given the reported internal concerns about data exposure, Meta has strong incentive to ship its promised encrypted Muse variant sooner rather than later, likely before year-end.
Regulatory attention follows quickly. An agent with access to a user’s Facebook, Instagram, and WhatsApp activity, plus email and payment details, is exactly the kind of cross-app data reach that has drawn scrutiny from privacy regulators in the past, and Muse’s US-first, adults-only launch looks like Meta getting ahead of that pressure rather than avoiding it entirely.

The Bottom Line

Muse is Meta’s clearest statement yet that it wants to compete on AI products, not just AI models. The company spent years being measured against OpenAI and Anthropic on raw model quality through Llama, and largely lost that argument in the eyes of many analysts. With Muse, Meta is changing the terms of the competition entirely, betting that distribution, pricing, and deep integration into apps people already use every day matter more than winning a benchmark leaderboard. Whether that bet pays off depends less on how smart Muse Spark is and more on whether millions of ordinary users decide they trust an AI agent with their inbox, their calendar, and their payment card.

Frequently Asked Questions

What is Meta Muse?

Muse is Meta’s personal AI agent, announced September 8, 2026. Meta describes it as a secure, private agent that handles tasks like emails, travel booking, form-filling, bill negotiation, and long-term planning on a user’s behalf, rather than just answering chat questions.

How much does Meta Muse cost?

Muse has a free tier plus two paid plans, reportedly Power at $20 a month and Maximum at $100 a month. Zuckerberg said the free tier covers usage up to 100 million tokens per week.

Is Meta Muse available outside the United States?

Not at launch. Reports describe Muse as available only in the United States when it launched on September 8, 2026, with no confirmed date yet for international expansion.

Do I need a credit card to use the free tier?

Reports say yes. Muse reportedly requires a payment card on file to start using the service, even on the free tier, which is a stricter requirement than many free AI chat products currently ask for.

What model powers Meta Muse?

Muse runs on Meta’s Muse Spark model family, developed under Meta Superintelligence Labs. Muse Spark 1.3, an earlier release in the same family, focused on cutting tool-call counts and token usage in agentic tasks.

Can I stop Muse from using my data to train Meta’s AI models?

Reports indicate users can opt out of having their Muse interactions used for AI model training. Meta has also said it plans to introduce an encrypted version of Muse later in 2026.

How does Muse compare to ChatGPT Atlas or Grok 4.5?

Pricing-wise, Muse’s $20 to $100 tiers sit above Grok 4.5’s reported $2 to $6 tiers and roughly in line with GPT-6 Astra’s reported $10 to $50 tiers. Muse’s main differentiator is distribution, since it’s built directly into WhatsApp, Facebook, and Instagram rather than launching as a standalone product.

Is there advertising inside Muse?

Not at launch, according to reports. Meta is reportedly exploring commerce-related features inside Muse over time, though no advertising has been confirmed for the initial rollout.

Related Coverage