Meta introduced Muse on September 8, 2026, calling it a personal AI agent built to run errands, manage schedules, and carry out multi-step tasks across a user’s digital life. The rollout, confirmed by TechCrunch, CNET, and SiliconANGLE, arrives in the US first, restricted to users 18 and older, and pushes Meta further into a race against OpenAI, Google, and Microsoft over who controls the AI agent that actually acts on a person’s behalf rather than just answering questions.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Meta actually shipped with Muse
Muse is not another chatbot bolted onto the Meta AI app. According to Reuters’ reporting on the September 2026 launch, it is a task-executing agent that connects to at least six categories of a user’s digital life — email, calendar, payments, health, shopping, and smart home — then works through multi-step goals with minimal supervision. Meta’s own framing, cited in reporting on the launch, describes the product this way: “Once a person shares a goal with Muse, it helps them develop a personalized plan and coordinate their time and resources, then advances the work on its own,” and adds that “it can open a browser, fill out forms, and negotiate on their behalf,” per WTOP’s coverage of the launch statement.
Meta’s official positioning statement goes further than describing what Muse does step by step. The company called Muse “the world’s first personal AI agent built for everyone,” a description that frames the September 8 launch as a category claim rather than a routine feature update, aimed squarely at everyday tasks like sending emails, booking travel, and pulling information from a user’s connected apps. That “for everyone” framing lines up with the generous free tier Meta built into the launch, covered in detail below, rather than positioning Muse as a premium add-on reserved for existing power users.
That distinction matters. Most consumer AI products in 2026 still operate as advisors: they draft an email, suggest a plan, or summarize a document, and a human executes the final step. Muse is pitched as doing the execution itself, inside what Meta describes as a dedicated, secure virtual machine running in its cloud, according to reporting from Bastille Post and Axios. That VM reportedly houses both the agent and a user’s connected data, with a visible browser window so people can watch the agent work rather than trust it blindly. Meta detailed that VM-based architecture as part of the September 8, 2026 launch, and Reuters reports the company plans to add an encrypted Muse mode later this year, targeting a rollout by December 2026 to further wall off the agent’s access to sensitive data.
Meta CEO Mark Zuckerberg framed the ambition in sweeping terms. In a quote surfaced by WTOP’s report on the launch, he said: “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.” That is a much bigger claim than “helps you write faster,” and it puts Muse in direct competition with the agentic ambitions Google, OpenAI, and Microsoft have all signaled for their own assistants.
Availability, platforms, and the age gate
Muse launched in the US only, and Meta capped access at 18 and older, a restriction confirmed independently by the Independent, Yahoo Finance, and Bastille Post. That age gate is notable given Meta’s history of scrutiny over how its products handle minors, and it suggests the company is trying to avoid a repeat of the content-moderation and safety controversies that have dogged its social apps.
At launch, Meta made Muse accessible across four surfaces — a dedicated iOS and Android app, a standalone web destination at muse.ai, and WhatsApp — according to TechCrunch, CNET, SiliconANGLE, and the Associated Press, which also confirmed the US-only, 18-and-older restriction. That WhatsApp integration is a meaningful distribution advantage: Meta does not need to convince people to download a new app when it can drop an agent into a messaging surface with billions of existing users. Meta has also said it plans to extend Muse to its AI glasses lineup, though TechCrunch and Axios both describe that as “coming soon” rather than available now, so treat glasses support as a roadmap item rather than a shipped feature.
Pricing: a generous free tier, then a steep jump
Muse launches with a free tier and two paid plans. Reporting from TechCrunch and Axios, and confirmed directly to CNBC by Meta, lists the three price levels as free, Power at $20 a month, and Maximum at $100 a month, a structure Reuters also reported following the September 8, 2026 launch. On the free side, Zuckerberg has said Muse will be usable at no cost up to 100 million tokens per week per user, a figure reported by CNET. That is a large allowance for a consumer product, and it signals Meta is prioritizing adoption over near-term monetization, similar to how it subsidized Llama model access and Meta AI chat before now.
The jump from free to $20, and then to $100, is steep compared to how most consumer AI subscriptions are structured today. It puts Muse’s top tier well above a typical single-app AI subscription and closer to what businesses pay for productivity software bundles. That pricing shape suggests Meta expects its heaviest users, people running Muse continuously across shopping, scheduling, and household admin, to look more like power users of a personal assistant service than casual chatbot users.
| Tier | Monthly Price | Usage Allowance | Access Points |
|---|---|---|---|
| Free | $0 | Up to 100 million tokens/week (per Zuckerberg, via CNET) | iOS, Android, muse.ai, WhatsApp |
| Power | $20 | Higher usage ceiling than free tier | iOS, Android, muse.ai, WhatsApp |
| Maximum | $100 | Highest usage ceiling reported | iOS, Android, muse.ai, WhatsApp |
Meta has not published exact token or task-volume ceilings for the Power and Maximum tiers in the coverage available at launch, so treat the specific usage caps above as directionally accurate rather than exact figures until Meta publishes a formal pricing page with hard numbers.
The pricing math: what $20 and $100 actually buy
Line up the three tiers and the structure looks less like a typical software price ladder and more like a usage meter. Free covers up to 100 million tokens a week, Power costs $20 a month, and Maximum costs $100 a month, a five-times jump from Power to Maximum with no intermediate step in between. Reuters, covering the same launch, described the two paid tiers as built for people with heavier usage needs rather than as gateways to different features, which matches what Meta has said publicly: nothing in the launch materials suggests Power and Maximum unlock different capabilities, only more headroom within the same agent.
That is a meaningfully different model than the tiered-feature approach common across consumer software, where a higher price usually buys a new capability, not just more of the same one. By pricing Muse on usage volume alone, Meta is betting that its heaviest users, people running the agent continuously across shopping, scheduling, and household admin, will hit the free tier’s ceiling often enough to justify paying $20 or $100 a month, while the “for everyone” positioning keeps the free allowance generous enough that casual users never need to see a paywall at all. Whether that math works depends entirely on the exact token and task ceilings tied to each paid tier, and Meta has not published those figures yet.
How Muse relates to Muse Spark, Meta’s coding model
The naming overlap between Muse the consumer agent and Muse Spark 1.3, the agentic coding model Meta shipped on September 2, 2026, is not a coincidence. Muse the consumer agent runs on the latest generation of models developed under Meta’s chief AI officer, Alexandr Wang, a lineage that includes the Muse Spark model family, according to Axios’s separate reports on both launches. Muse Spark 1.3 itself is aimed at developers, first announced September 2, 2026 and made broadly available through Muse Code and Meta’s Model API on September 8, 2026, per MarkTechPost, for long-running coding and agentic workflows, not at consumers managing their calendars.
In practice, that means Meta built one underlying agentic model architecture and then shipped two different products on top of it, one for developers who want an agent that writes and executes code, and one for consumers who want an agent that books appointments and fills out forms. That is a similar playbook to how OpenAI has split ChatGPT’s consumer-facing agent features from Codex, its developer-focused coding agent, and how Google has separated Gemini’s consumer assistant from its developer tooling.
Six days, two launches: reading Meta’s release cadence
The gap between Meta’s two September announcements is worth pausing on. Muse Spark 1.3 shipped on September 2, 2026, aimed at developers building coding and agentic workflows, and Muse followed just six days later, on September 8, 2026, aimed at consumers. Shipping a developer-facing model and a consumer-facing agent built on the same underlying lineage within the same week is not the kind of thing that happens by accident at a company Meta’s size; it points to coordinated planning across Meta’s AI organization under chief AI officer Alexandr Wang, timed so that both technical and mainstream press covered Meta’s agent push within days of each other.
That sequencing also served a narrower purpose: it let Meta establish credibility with developers first, through Muse Spark 1.3’s coding and agentic workflow capabilities, before asking a much larger, less technical audience to trust an agent with their email and payment methods a week later. Compared with how OpenAI and Google have generally spaced out their developer and consumer agent announcements by weeks or months, Meta’s back-to-back rollout reads as a deliberate attempt to dominate the agentic AI news cycle for a full week rather than split attention across two separate moments.
The trust problem: letting an agent spend your money
TechCrunch’s launch coverage is framed around a direct question: will consumers trust an AI agent enough to let it act autonomously with their accounts and money? Meta’s answer, according to reporting from Agent Log and Bastille Post, is a permission model where Muse asks for explicit user approval before sensitive actions, specifically before sending a message on the user’s behalf or spending money. Users can also control what data and accounts the agent can see and access in the first place.
That approval gate is the same basic safety pattern other agentic AI products have converged on in 2026: broad autonomy for low-stakes actions like drafting a plan or browsing for information, and a hard stop requiring human sign-off before anything irreversible, like a purchase or a message sent to another person. Whether that pattern actually prevents costly mistakes at scale, once millions of people are running agents continuously, is the open question every company shipping an agent in 2026 is being asked, and Meta is no exception.
Where Muse fits in the broader agent race
Muse does not launch into an empty field. Meta’s own Hatch AI agent platform already exists as a separate agent effort, and rivals have been racing to ship browser and task agents throughout 2026. OpenAI, Google, and Perplexity have all shipped competing agent products, a landscape tech-insider.org has covered in a head-to-head comparison of Comet, Gemini Agent, and ChatGPT Atlas. Muse’s pitch, an agent that lives inside WhatsApp and a dedicated app rather than a browser extension, is a distinct distribution bet: instead of asking users to change how they browse the web, Meta is asking them to message an agent the way they’d message a person.
That distribution angle plays to Meta’s core strength. WhatsApp alone counts billions of monthly users worldwide, and folding an agent directly into that chat surface means Meta does not have to win a separate battle for app downloads or browser market share the way OpenAI and Perplexity do with their agent products. It is the same logic Meta has applied to Meta AI generally: put the assistant where the users already are, rather than building a new destination and hoping people show up.
| Agent Product | Company | Primary Interface | Consumer or Developer Focus |
|---|---|---|---|
| Muse | Meta | Dedicated app, muse.ai web, WhatsApp | Consumer |
| Muse Spark 1.3 | Meta | Muse Code, Model API | Developer |
| ChatGPT Atlas | OpenAI | Browser | Consumer |
| Gemini Agent | Browser, Gemini app | Consumer | |
| Comet | Perplexity | Browser | Consumer |
Historical context: from chatbots to agents in under three years
It is worth remembering how fast this shift happened. ChatGPT’s public launch in late 2022 introduced most consumers to conversational AI as a question-and-answer tool. Meta’s own Llama models followed a similar path, first as a research release, then as the engine behind a chat assistant folded into Instagram, WhatsApp, and Facebook. Muse represents the next step in that progression: instead of an assistant that talks, Meta is shipping one that acts, with direct access to a user’s calendar, payment methods, and connected apps.
That progression mirrors what happened across the industry more broadly in 2026. Coding assistants moved from autocomplete to autonomous agents that write, test, and ship code with limited supervision. Shopping assistants moved from product recommendations to agents that complete checkout on a user’s behalf, a trend tech-insider.org has tracked in its comparison of ChatGPT, Copilot, and Gemini as shopping agents. Muse applies that same execution-over-conversation shift to personal life management broadly, rather than to a single vertical like shopping or coding.
What could go wrong: the risks agentic products carry
Handing an AI agent access to email, payment methods, and a live browser session raises the same categories of risk that have shadowed every agentic AI launch in 2026: prompt injection attacks that trick an agent into taking unintended actions, mistaken purchases or bookings made on a user’s behalf, and the privacy exposure that comes from an agent reading through a person’s inbox and calendar to plan tasks. Meta’s approval-gate design, requiring explicit sign-off before spending money or sending messages, addresses the most obvious failure mode, but it does not eliminate risk from an agent that misreads context or acts on stale information.
There is also a regulatory dimension. Meta has faced repeated scrutiny in the US and Europe over how its products handle user data, and an agent that actively browses, negotiates, and transacts on a person’s behalf touches far more sensitive systems than a chat interface ever did. The 18-and-older age restriction at launch looks like a deliberate attempt to get ahead of that scrutiny rather than face it after the fact, given how much attention regulators have paid to Meta’s handling of younger users across its other products.
Market impact: what this means for Meta’s AI strategy
Muse arrives at a moment when Meta has been investing heavily in AI infrastructure and talent, including its restructured AI research effort under chief AI officer Alexandr Wang. Investors responded immediately: Meta’s stock rose 6% on September 8, 2026, the day of the Muse reveal, according to SiliconANGLE. Shipping a flagship consumer agent product gives Meta a visible answer to the question investors and the press have been asking all year: what does Meta’s massive AI spending actually produce for the average user, as opposed to advertisers or developers. A free tier generous enough to cover meaningful weekly usage, paired with premium tiers priced above typical consumer AI subscriptions, suggests Meta is betting on volume first and revenue second, the same approach it has taken with WhatsApp and Instagram in past product cycles.
For competitors, Muse’s WhatsApp integration is the detail worth watching most closely. If Meta can convert even a small fraction of WhatsApp’s user base into regular Muse users, it instantly has one of the largest agentic AI user bases in the world without needing to win a separate download or browser-adoption battle. That is a structural advantage OpenAI, Google, and Perplexity cannot replicate quickly, since none of them own a messaging platform at WhatsApp’s scale.
What it means for advertisers and merchants
Meta’s business has run on advertising for two decades, and an agent that shops, books, and pays on a user’s behalf sits close to that revenue engine. If Muse becomes the layer that decides which movie ticket vendor to use or which store to buy from, Meta gains a new kind of leverage over merchants that goes beyond showing them an ad: it can become the checkout path itself. None of the launch coverage from TechCrunch, CNET, or Axios details a merchant or advertiser monetization model for Muse yet, so this remains a reasonable inference about incentives rather than a confirmed plan, but it is the same dynamic that made Amazon’s product search and Google’s shopping tab strategically important to those companies.
That also raises a question for e-commerce brands: if agents like Muse increasingly complete purchases without a human clicking through a storefront, the traditional playbook of optimizing a product page for a human shopper starts to matter less than optimizing for what an AI agent can parse and act on. Retailers that have already adapted to agentic shopping traffic from ChatGPT and Copilot, a shift tech-insider.org detailed in its look at AI shopping agents, will likely face a similar adjustment with Muse.
What developers and IT teams should watch
For developers, the more immediate product is Muse Spark 1.3 rather than the consumer Muse agent, since that is the model Meta has opened up through Muse Code and its Model API for building agentic and coding workflows. IT and security teams evaluating whether to allow Muse inside their organizations, for instance if employees connect work email or calendars to a personal agent, should treat it the same way they would any new SaaS integration: check what data it can access, whether it can take irreversible actions like sending messages or making purchases, and whether an admin can audit or revoke that access. Meta’s stated approval-gate model, requiring sign-off before spending money or sending a message, is a reasonable baseline, but enterprise security teams have historically wanted stronger, centrally managed controls than a per-user consumer approval prompt provides.
Predictions: what happens next
Based on the launch details and the pattern of prior Meta AI rollouts, a few things are likely to follow. First, expect Meta to expand Muse beyond the US within the next several months, following the same phased rollout pattern it used for Meta AI chat features in WhatsApp and Instagram. Second, expect scrutiny over data privacy and permission scope to intensify once Muse has a larger user base, particularly around what happens to the data an agent collects while browsing and transacting on a user’s behalf. Third, expect Meta’s AI glasses integration to arrive as a headline feature within the next couple of product cycles, since Meta has already signaled it as a near-term roadmap item.
Fourth, expect rival agent products from OpenAI, Google, and Perplexity to lean harder into messaging-platform distribution in response, since Muse’s WhatsApp integration exposes how much of an advantage owning a chat surface provides. Fifth, expect Meta to eventually publish more granular usage limits and task-volume caps for the Power and Maximum tiers, since the current reporting describes pricing but not the exact ceilings tied to each plan, a gap Meta will likely close once the product exits its initial rollout phase.
What Meta still hasn’t disclosed, a week after launch
More than a week after the September 8, 2026 launch, several obvious data points remain missing from Meta’s public statements. There is no published figure for how many people have signed up for Muse, no breakdown of how many of those users sit on the free tier versus Power or Maximum, and no confirmed date for expansion beyond the US. That silence is not unusual this early in a rollout, but it does mean that every claim about Muse’s traction right now, including this article’s read on its distribution advantages through WhatsApp, is a projection based on Meta’s stated strategy rather than a measurement of actual adoption.
The same gap applies to the hard usage ceilings for Power and Maximum. Meta has said the two paid tiers exist for heavier users, per Reuters, but has not said how many tokens or tasks $20 or $100 actually buys, which makes it impossible right now to calculate a real price-per-task comparison against other agentic products. Readers evaluating whether to pay for Muse should treat the current pricing as a starting point Meta can and likely will refine once it has usage data from the free tier at scale, rather than as a final, fixed structure.
Frequently asked questions
What is Meta’s Muse?
Muse is a personal AI agent Meta launched on September 8, 2026, designed to carry out tasks like scheduling, shopping, and form-filling on a user’s behalf rather than just answering questions.
Is Muse the same as Muse Spark 1.3?
No. Muse Spark 1.3 is a developer-focused coding and agentic model Meta released on September 2, 2026, distributed through Muse Code and Meta’s Model API. Muse, the consumer agent, runs on the same underlying model family but is a separate, consumer-facing product.
How much does Muse cost?
Muse offers a free tier, reported by Mark Zuckerberg as usable up to 100 million tokens a week, plus two paid tiers: Power at $20 a month and Maximum at $100 a month.
Why does Muse jump straight from free to $20 to $100?
Meta has not detailed exact usage ceilings, but Reuters reported the two paid tiers are aimed at people with heavier usage needs rather than different features, meaning Power and Maximum buy more headroom within the same agent, not new capabilities.
Where is Muse available?
At launch, Muse is available only in the US, for users 18 and older, through a dedicated iOS and Android app, the muse.ai website, and WhatsApp.
Can Muse spend money without permission?
According to Meta’s launch statements, Muse asks for user approval before sensitive actions, including spending money or sending a message, and users control what data and accounts the agent can access.
Will Muse work on Meta’s AI glasses?
Meta has said support for its AI glasses is coming, but it was not available at the September 8, 2026 launch. Treat it as a planned feature rather than a current one.
How does Muse compare to ChatGPT Atlas or Gemini Agent?
All three are agentic AI products aimed at completing tasks rather than just chatting, but Muse’s primary distribution is through a dedicated app and WhatsApp, while Atlas and Gemini Agent are built primarily as browser-based agents.
Who leads Meta’s AI efforts behind Muse?
Muse runs on models developed under Alexandr Wang, Meta’s chief AI officer, according to Axios’s reporting on the launch.
Related Coverage
- Meta Ships Muse Spark 1.3: 20% Fewer Tool Calls [2026]
- GPT-6 Astra Launches: $10/M Tokens, How to Try It [2026]
- OpenAI Agent Swarm Hit 7 Sites, Firm Denies Coverup [2026]
- Gemini 3.8 Flash Cyber Launches: 2.6x Better Chrome Patches [2026]
- Astra Hits 100% on ExploitBench, Alarms White House [2026]
