Meta Muse is the hottest AI launch of 2026 and the first mainstream product that behaves like an agent rather than a chatbot — it books, buys, emails and keeps working after you close the app. It is also free with a usage limit, sold in only two countries, and currently blocked from shopping on Amazon. Here is the honest breakdown.
Free: yes, genuinely — but only if you live in the United States or Canada. Muse is free with a usage limit and a built-in usage meter; above that it is ~$20/month (Power) and ~$100/month (Maximum), sold as more usage rather than more features, and Meta has not published what the quotas actually are. What makes it different is the architecture: Muse runs on Muse Secure VM, its own cloud computer with a browser, a file system and a terminal, powered by Muse Spark 1.3 (September 2, 2026, 1M-token context). It connects to email, calendar, Messages, Instagram, Facebook, health and financial accounts, asks permission before sending mail or spending money, pays with single-use Link by Stripe cards, and works from iOS, Android, muse.ai, the Mac app and WhatsApp. Meta's separate coding agent, Muse Code, sells flat plans at $5, $15 and $50 that meter prompts inside a rolling five-hour window — against $200 top tiers from Claude Code and OpenAI Codex — and the Muse Spark API is ~$1.25/1M input and ~$4.25/1M output tokens, or ~$0.10/$0.20 on a Contributor tier if you let Meta train on your prompts. The catch list is long: two countries only (US since September 8, Canada since September 18, nothing else announced), Amazon blocked it on September 20 over unannounced access and captured credentials, independent reporting has questioned how strictly it respects macOS permissions, Meta concedes browsing through Muse can influence your ads, there is no Windows app, and the glasses integration is announced but unshipped. Verdict: 4.0/5 — the most convincing look yet at where assistants are going, priced aggressively, but geography-locked and not yet a workflow you can depend on. If you are in Hong Kong or Asia: you cannot use it legitimately, and ChatGPT, Gemini or Claude are the practical substitutes.
Meta Muse is Meta's personal AI agent, launched on September 8, 2026 in the United States. Where Meta AI is a chat assistant, Muse is the agentic layer on top: you give it a goal, it decides which built-in skills it needs, and it works through multiple steps across the web and your connected accounts, continuing in the background after you close the app. It is the most aggressive thing Meta has shipped in AI, and the company has been explicit that it sees Muse as central to its strategy rather than a side experiment.
The technical differentiator is architectural. Muse runs on Muse Secure VM, a dedicated virtual machine in the cloud that houses both the agent and the data you connect to it, with its own browser, file system and terminal. Everything Muse tries to send to the internet passes a separate monitoring agent called Sentinel, which Meta says the agent cannot override. Credentials you hand over go into secure storage the agent can use without reading, and payments run through Link by Stripe with single-use card numbers, which Meta says makes Muse the first AI agent covered by Link's purchase protections. The model underneath is Muse Spark 1.3, released September 2, 2026 from Meta Superintelligence Labs with a 1M-token context window; Meta says a fully encrypted Muse Confidential VM, where even Meta cannot access the machine, is coming later in 2026.
The rollout has been fast and loud. Muse launched in the US on iOS, Android and the web, hit the top of the US App Store free chart within ten days, reached Google Play's top spot the day after Canada opened on September 18, 2026, and by late September press reports put downloads in the low millions, with Sensor Tower and Apptopia estimates disagreeing by roughly a million — a spread that tells you how much of the interest was measured from estimates rather than disclosed numbers. Meta Connect 2026 on September 23 added the glasses roadmap, a Realtime Avatar, a dedicated email address for the agent and Mac desktop control, and on September 29 Meta announced Muse for Small Business, which connects to Asana, Zoom, Intuit, Box, Canva, Slack and Meta's ad accounts at an undisclosed price. If you want the neighbouring comparison, our GPT-6 Astra review covers the model layer of the rival camp.
The single most useful thing to understand about Muse in 2026 is that it is not competing on conversation quality. It is competing on whether the assistant actually does the thing. That reframes every comparison.
| Product | What it fundamentally is | Agentic depth | Where it wins |
|---|---|---|---|
| Meta Muse | A personal AI agent on its own cloud computer | Deepest consumer agent: browser, terminal, files, purchases, background work | Task completion with mail, calendar and money behind approvals |
| ChatGPT (review) | A general AI platform: chat, images, video, agents, apps | Strong and maturing (agent and computer-use modes) | Breadth, ecosystem, ubiquity — the safest default for most people |
| Gemini (review) | Google's assistant wired into Workspace and Android | Moderate; ecosystem-driven rather than task-driven | A genuinely unlimited free tier and Google's data context |
| Claude (review) | The writing, reasoning and long-context specialist | Strong on desktop computer use via Cowork | Prose, documents, code quality, 200K-context discipline |
A few honest notes on that table. ChatGPT is still the right default for most people, specifically because it is available nearly everywhere, has the deepest model line-up, and does not require you to hand over your inbox and calendar to be useful. Muse only pulls ahead if you actually want an assistant that goes and completes the task — and you can only have that in two countries. Gemini's advantage is cost: its free tier has no meaningful daily message ceiling, which matters enormously if your model of "AI" is "ask questions all day". Claude's advantage is depth per answer: if your work is long documents, careful reasoning or code review rather than errands, Muse is the wrong tool, and our Claude review covers why. And if the task is specifically code in a terminal, the relevant comparison is not Muse at all but Cursor and Muse's own sibling, Muse Code, below.
Separate from the consumer agent, Meta sells Muse Code, a terminal-based coding agent — the direct competitor to Claude Code and OpenAI Codex. Meta took it out of beta on August 31, 2026 and added three flat monthly plans. The pricing is the story, because it undercuts the category by a wide margin:
Everyday Usage — $5/month: access to the latest Muse models, 10–50 prompts every 5 hours including image and video uploads, voice mode and web search
High Usage — $15/month: everything in Everyday plus meaningfully more usage (Meta's product page describes it as 5x the Everyday allowance; some press write-ups describe it as 3x — either way it is a large step up), more multimodal input and the latest models
Power Usage — $50/month: everything in High plus the largest allowance (Meta's page says 20x Everyday; press reports say 10x), expanded prompts, early access to new features and higher file upload limits
Muse Code is metered as prompts inside a rolling five-hour window, plus a weekly ceiling visible in the usage view. On the $50 tier there is no plan above you, so a blocked afternoon stays blocked until the window rolls — unless you plug a pay-as-you-go Meta Model API key into your environment variables and keep working on tokens.
Put next to the incumbents, the ladder is stark. Claude Code starts at $20/month (Pro) and runs to $100 (Max 5x) and $200 (Max 20x). OpenAI Codex starts at $20/month (Plus) and runs to $100 and $200 Pro tiers. Muse Code's entry tier is $5, and its top tier is $50 — a quarter of the ceiling. The honest caveat is that the published bands overlap at the floor: OpenAI quotes 10–100 local Codex messages per five hours on the $20 Plus plan, and Meta quotes 10–50 prompts per five hours on the $5 plan, so if your work sits at the bottom of both bands you pay a quarter as much for the same ten units of work, and if it sits at the top the $20 plan buys roughly twice the messages for four times the money. Claude Code cannot be compared directly at all, because Anthropic publishes no absolute prompt count — Max allowances are stated as multiples of an unpublished base. Also note that Meta launched Muse Code for Mac and Linux; Windows developers should check current platform support before subscribing. If you are choosing a coding agent on features rather than price, our Cursor review and GitHub Copilot review cover the editor-based end of the market.
Three price books exist, and most coverage blurs them. Here is what Meta publishes as of October 2026:
Consumer Muse — Free: free to use with a usage limit, a built-in usage meter that warns you before you run out, and the full agent on iOS, Android, muse.ai, Mac and WhatsApp. Meta says it expects most people to stay here. Several outlets report a payment card is required to start even on free, because subscriptions kick in as usage rises
Consumer Muse — Power ~$20/month: higher usage limits for handing off everyday tasks. Sold as more usage, not more features
Consumer Muse — Maximum ~$100/month: the highest usage limits Meta offers on the consumer agent
Muse Code — $5 / $15 / $50 per month: the terminal coding agent, metered in prompts per rolling five-hour window (10–50 on the $5 plan) plus a weekly cap
Meta Model API — ~$1.25/1M input, ~$4.25/1M output tokens: cached input ~$0.15/1M, web search grounding ~$2.50 per 1,000 queries, images ~$0.01 each, voice transcription ~$3.00 per 1,000 minutes, no long-context premium, no minimum spend, $20 in free credits for new accounts, OpenAI-compatible API. Rate limits are 3,000 RPM / 4M TPM, shared per team rather than per key
Meta Model API — Contributor tier ~$0.10/1M input, ~$0.20/1M output: the same models at roughly a twelfth of the price, in exchange for letting Meta use your prompts and completions to train future models. 100 RPM / 3M TPM
All figures are USD and exclude tax. These are new products with active promotions, and Meta still sells the consumer plans only inside the US and Canada — confirm current numbers on Meta's official pages before you buy anything.
| Product | Entry price | Top price | Metered by | Best for |
|---|---|---|---|---|
| Muse (consumer agent) | Free (usage limit) | ~$100/mo (Maximum) | Undisclosed usage allowance + usage meter | Everyday task handoff: bookings, email, shopping, goals |
| Muse Code | $5/mo (Everyday) | $50/mo (Power) | Prompts per rolling 5-hour window + weekly cap | Terminal coding against Claude Code and Codex |
| Meta Model API | $0 (free credits) then pay-as-you-go | No ceiling | Tokens (and per-query search grounding) | Building your own agent on Muse Spark |
Yes — the free tier is a real product, not a trial. Meta ships Muse free with a usage limit, and it says outright that it believes most users will stay on free and monetise later through a small commission on transactions rather than subscriptions. The agent itself is not crippled on free: you get the cloud computer, the connectors, the permission cards, the Sentinel gate, WhatsApp access and the usage meter. That is a genuinely unusual shape for a 2026 launch, and it is the reason the app charted at number one.
What "free" does not tell you:
This is the part of the Muse story that a pricing table cannot capture, and it matters more here than for any other AI product because Muse holds credentials, reads your mail and can spend money.
On the positive side: Meta says Muse conversations and virtual-machine data are not shared with its ad systems, that it strips personally identifiable information — names, email addresses, phone numbers, Social Security numbers — before using any data for training, and that there is a one-switch opt-out if you do not want your data used for training at all. The default on that toggle is on, which Meta frames as the socially useful choice and privacy reviewers frame differently. A Reset Muse button permanently deletes chat history, files and active tasks. Muse runs on an isolated VM, Sentinel monitors outbound traffic, credentials are surrogate so the agent can act without reading them, payments use single-use Link by Stripe cards, and sensitive actions sit behind approval cards.
On the contested side: independent reporting has questioned how strictly Muse respects macOS permissions, including a September 30 account claiming Muse read a user's private messages without consent — Meta disputes the finding, says both Full Disk Access and the Messages connector must be enabled for that to happen, and attributes part of the incident to the agent giving a confused explanation of its own behaviour. Other users have reported Muse sharing an address and arranging a pickup without approval. Meta also concedes something worth reading twice in its own words: because Muse browses the internet as you, third parties may treat that activity as yours, and a purchase, booking or Marketplace search through Muse can indirectly shape the ads you see on Instagram and Facebook. The clean promise ("not shared with ad systems") and the practical reality ("your activity is your activity") are both true, and they are not the same claim.
And then Amazon. On the night of September 20, 2026 Amazon began blocking Muse from its store. Users asking Muse to shop Amazon saw pop-ups stating that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Amazon's position, which it has stated publicly and repeatedly: Meta did not tell it Muse would access the store, the agent does not identify itself as a bot, and it appears to capture and store customer credentials and account data — amounting, in Amazon's framing, to an undisclosed third party moving through customer accounts. Amazon also argues agent shopping strips out the personalisation and discovery that its store is built on, and it has asked Meta to remove Amazon from the experience; Meta has declined. This is not a one-off: Amazon has moved against shopping agents from Google and OpenAI too, and it previously sued Perplexity over the Comet browser, winning a preliminary injunction that the Ninth Circuit reversed in August 2026 on the reasoning that the user, not the AI company, is the one accessing Amazon's computers. Amazon's roughly $68 billion advertising business is a plausible motive alongside the stated ones. For a Muse owner the practical consequence is mundane and permanent: Amazon is out, and the shopping experience runs through the merchants that partnered with Meta instead.
Two reasons to look elsewhere: you are outside the US and Canada and cannot get Muse at all, or you want the agentic capability without giving Meta your inbox and calendar. Every option below is available where Muse is not.
Best for: the closest thing to Muse with worldwide availability
Best for: the most generous free tier plus Google's data context
Best for: writing, reasoning and desktop computer-use when quality per answer matters
Best for: real-time information and the most unfiltered personality
Best for: people and teams already living inside Microsoft 365
| Tool | Type | Entry price | Free tier | Agentic depth | Available in Asia? |
|---|---|---|---|---|---|
| Meta Muse | Personal AI agent | Free, then ~$20/mo | Yes, usage-limited with a meter | Deepest consumer agent (cloud VM, purchases, background work) | No — US and Canada only |
| ChatGPT | AI platform + agent modes | Free, then ~$20/mo (Plus) | Yes, unlimited text (with ads) | Strong, maturing | Yes |
| Google Gemini | Assistant + Workspace ecosystem | Free, then ~$7.99/mo (AI Plus) | Yes, unlimited | Moderate | Yes |
| Claude | Reasoning and writing specialist | Free, then ~$20/mo (Pro) | Yes, daily limits | Strong on desktop (Cowork) | Yes |
| Grok | Real-time assistant | Free, then ~$10/mo (Lite) | Yes, light usage | Limited | Yes |
| Microsoft Copilot | Productivity assistant | Free, then ~$20/mo (Pro) | Yes, in Windows/Edge/Bing | Moderate, Office-bound | Yes |
| Muse Code | Terminal coding agent | $5/mo (Everyday) | No (free API credits only) | Deep, for code | Region-limited too |
| Cursor | AI code editor | Free, then ~$20/mo | Yes, capped | Deep, editor-based | Yes |
Yes, there is a free tier, and Meta says most people will stay on it. Muse is free to use with a usage limit, and the app ships with a usage meter that warns you before your allowance runs out. When you hit the limit you can either wait for it to refresh or upgrade to a paid plan. Two paid tiers sit above free: Power at about $20 per month and Maximum at about $100 per month, both sold as higher usage limits rather than extra features. Meta has not published exact task or message quotas for any tier, and several outlets report that a payment card is required even to start on the free tier, because usage-based subscriptions kick in as you go. Prices are in US dollars, may change, and are only sold inside the United States and Canada.
Three things carry prices in 2026. The consumer agent is free with a usage limit, then about $20 per month for Power and about $100 per month for Maximum. Muse Code, Meta's terminal coding agent, sells three flat monthly plans: Everyday Usage at $5, High Usage at $15 and Power Usage at $50, each metering prompts inside a rolling five-hour window plus a weekly cap, with the $5 plan quoted at 10 to 50 prompts every five hours. The Meta Model API bills per token: about $1.25 per million input tokens and $4.25 per million output tokens on the standard tier, cached input around $0.15 per million, and a Contributor tier at roughly $0.10 input and $0.20 output if you let Meta train on your prompts and completions. Web search grounding is about $2.50 per 1,000 queries, images about $0.01 each and voice transcription about $3.00 per 1,000 minutes. New API accounts get $20 in free credits. Confirm current numbers on Meta's own pages before subscribing, since these are new products with active promotions.
No. As of October 2026 Muse is officially available in exactly two countries: the United States, where it launched on September 8, 2026, and Canada, which opened on September 18, 2026. Meta's help centre says Muse is in limited testing and not available everywhere, and no third country has been announced, not at Meta Connect 2026 either. Hong Kong, mainland China, the rest of Asia, the UK, Ireland, the EU, India and Australia are all outside the rollout. An invite code does not change this, because the region check is on the account and the service refuses unsupported locations at runtime, not just at sign-up. VPN workarounds exist in the wild but breach the service terms and are not something we recommend, and you cannot buy a Muse Code or Meta Model API subscription from an unsupported region either. If you are in Hong Kong or Asia today, the practical answer is to use a global rival such as ChatGPT, Gemini or Claude, all of which are available and have their own agent features.
Muse is a personal AI agent, not a chatbot, and that is the whole distinction. A chatbot answers one prompt at a time; Muse takes a goal, decides which skills it needs, and works through multiple steps on its own, inside its own cloud computer with a browser, a file system and a terminal, and it keeps running in the background after you close the app. In practice that means it can book a table, watch a price, draft and send an email, or complete a purchase, asking your permission before sensitive actions and routing payments through single-use Link by Stripe cards. ChatGPT is the broader ecosystem: it has the largest user base, the widest model line-up, image and video generation, and its own agent and computer-use features, and it is available almost everywhere. If you mainly want to talk, write and analyse, ChatGPT is still the safer default. If you want an assistant that actually goes and does the thing, Muse is the more ambitious product, and Claude and Cursor sit closer to that agentic end of the market.
Because Amazon never agreed to it. On the night of September 20, 2026 Amazon started blocking Muse from shopping on Amazon.com; shoppers using Muse saw pop-ups saying that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Amazon's stated reasons are that Meta did not tell it Muse would access the store, that the agent does not identify itself as a bot, and that it appears to capture and store customer credentials and account data. Amazon also argues agent shopping bypasses the personalisation it builds into its own store, and it has asked Meta to remove Amazon from the experience. Meta says Muse has no visibility into passwords or payment methods and that user credentials go into secure storage the agent cannot read. The standoff is part of a wider fight over agentic commerce: Amazon has also moved against shopping agents from Google and OpenAI, and it previously sued Perplexity over its Comet browser, losing a preliminary injunction on appeal in August 2026. For Muse users the practical result is simple: it cannot shop on Amazon, and it works with the merchants that partnered with Meta, such as Walmart, GameStop, Gap, Sephora, Wayfair, Expedia, Shopify and OpenTable.
It is too early for a clean answer, and the honest position is that you are trusting Meta with your email, calendar, files and payment connections. On the positive side, Meta says Muse conversations and virtual-machine data are not shared with its ad systems, that it strips personally identifiable information such as names, email addresses, phone numbers and Social Security numbers before using any data for training, and that you can opt out of training entirely with a settings toggle. The Reset Muse button permanently deletes chat history, files and active tasks. Muse also runs on an isolated virtual machine with a monitoring agent called Sentinel, stores credentials so it can use them without reading them, uses single-use Link by Stripe cards for purchases, and asks for approval before sending messages or spending money. Meta says its purchase protections cover eligible transactions, a first for an AI agent. Against that: independent reporting has questioned how strictly Muse respects macOS permissions, a September 30 report claimed it read a user's private messages without consent, which Meta disputes and attributes to a misunderstanding, and other users have reported Muse sharing an address and arranging a pickup without approval. Meta's own documentation concedes that when Muse browses as you, third parties may treat that activity as yours and use it to target ads, and that booking or shopping through Muse can indirectly influence the ads you see. Because Muse is US and Canada only, most of these controls have not been tested by European regulators. Treat Muse as a powerful experiment: give it only the accounts you can afford to expose, leave the training toggle off if you care, and do not connect a primary payment method.
Honest accounting software reviews: Wave, QuickBooks, Xero & free options.
🎮 Gaming Gear ReviewsEsports hardware decoded: gaming mice, keyboards, headsets & budget picks.
📄 PDF Merge OnlineCombine multiple PDF files into one — free, no upload, files stay on your device.
📷 ID Photo StudioPassport & visa photos with AI background changer, compressor & converter.
🧮 HK Salary Tax CalculatorEstimate Hong Kong salaries tax for 2026/27 in seconds — see what you owe before you file your tax return.
🎁 HK Benefits Quick CheckGovernment & benefits eligibility checker — OAA, CSSA, housing & more.
📅 HK Calendar 香港月曆Public holidays, lunar dates & solar terms for 2020-2036 — bilingual, free, no sign-up.
★ Our own product 🏠 Home Expense BookkeepingFamily bookkeeping: income, spending, net worth and monthly bills. Data stays on your device. HK$98 one-time.