Is Meta's Muse Really the Biggest AI Launch Since ChatGPT? What Muse Wonder Does, Costs and Still Has to Prove
Artificial IntelligenceMeta MuseOctober 2026
Is Meta’s Muse Really the Biggest AI Launch Since ChatGPT? What Muse Wonder Does, Costs and Still Has to Prove
Meta released Muse Wonder, its first frontier AI model, on 12 October 2026, a month after its Muse agent app arrived on phones. The launch is Meta’s most important in AI so far. Whether it is the biggest since ChatGPT depends on numbers that are not in yet.
Over the past few days, a phrase has spread across tech feeds and video titles: the biggest launch since ChatGPT. It refers to Meta’s Muse, and in particular to Muse Wonder, the model Meta’s Superintelligence Labs released on Monday 12 October. Meta says Wonder reaches the frontier of AI capability, scoring alongside the best models from OpenAI and Anthropic on several tests.
That is a big claim, and so is the headline. ChatGPT changed how hundreds of millions of people use computers within months of its release in November 2022. To match that, Muse would need not just a strong model but a product that ordinary people pick up and keep using. This article sets out what Meta has actually launched, what it costs, how the model compares on the numbers available, and what would have to happen for the comparison with ChatGPT to hold.
What We Know
- Meta released Muse Wonder and Muse Wonder Full on 12 October 2026, describing them as its most capable models.
- Wonder is available through the Meta Model API and the Muse Code coding agent; Full is limited to higher Muse Code plans.
- The model card lists a 1 million token context window and text, image, video and audio inputs.
- The Muse agent app launched on iOS and Android on 8 September 2026 and on iPad on 7 October.
What Is Still Unclear
- When Wonder will power the Muse app for all users, not just subscribers.
- How Wonder performs on independent tests that Meta did not run or choose.
- Prices and availability in India and other markets outside the US.
- How many people pay for Muse, and how many keep using it after the first month.
Three Things Called Muse, and Why the Names Matter
Much of the confusion around this launch comes from the name. “Muse” now refers to a family of products, and the headline claim depends on which one you mean. Meta first used it for a model, then for an app, and now for a more powerful model inside both.
The models: Spark, Wonder and Wonder Full
Muse Spark, announced on 8 April 2026, was the first model from Meta Superintelligence Labs. TechCrunch describes it as lightweight and fast. Muse Wonder and Muse Wonder Full, released on 12 October, are larger and, by Meta’s account, far more capable, particularly at long, multi-step tasks. Meta calls Wonder “the next step on our scaling ladder and a major leap toward personal superintelligence.”
The app: Muse, a personal AI agent
Muse, launched on 8 September, is an assistant that can act, not just answer. Users connect accounts, which Meta calls connectors, so it can track email, meetings and bills, book reservations, order groceries and make purchases. TechCrunch lists connectors including Asana, Canva, Dropbox, Figma, GitHub, Notion, QuickBooks and Zoom, and retail partners including Best Buy, Gap, Sephora, Walmart and Wayfair.
The coding tool: Muse Code
Muse Code is a terminal-based coding agent for developers, with its own subscription plans. It is the first place Wonder Full is available to anyone outside Meta, which tells you who Meta expects to test the model hardest in the first weeks: developers rather than everyday users.
Where Wonder Sits Against the Leaders
Meta’s published comparisons focus on coding. On DeepSWE 1.1, which tests whether a model can resolve real software engineering tasks, Meta reports 79.0 for Wonder Full and 78.8 for Wonder, ahead of Claude Opus 5.5 at 74.2, GPT-6 Astra at 74.1 and Claude Fable 5.1 at 67.4. On SWE-Marathon, which tests longer tasks, the gap is wider.
TechCrunch reported more cautiously that Wonder and Wonder Full score “comparably” to Fable 5.1 and GPT-6 Astra overall, and that Wonder scores higher than rival models on MemoryBench, a test of how well a model builds and uses memory over time. The model card also lists tests in law, finance and health, but does not publish numeric scores for them.
What You Pay, Plan by Plan
Meta is selling Muse three ways: a consumer app with a free tier and paid plans, a set of Muse Code plans for developers, and a pay-as-you-go API. The consumer and developer subscriptions are separate.
The consumer Muse app launched with a free tier and paid plans at $20 and $100 a month, according to 24/7 Wall St. Meta says Wonder will reach Muse subscribers first. On the API, Wonder costs $4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. TechCrunch notes that this is roughly in line with GPT-6 Sol and Claude Opus 5.5. Wonder Full is not offered through the API at all.
The privacy setting to check
Muse Code subscriptions default to a “Contributor” mode, which lets Meta train on prompts, code, files and responses in exchange for more usage, according to Online Tech Tips’ reading of Meta’s terms. It can be switched off. Developers working on client or proprietary code should check this before their first session.
The Adoption Test: Muse Against ChatGPT’s First Months
“Biggest launch” is ultimately about people, not benchmarks. ChatGPT, released on 30 November 2022, was widely reported to have passed 1 million users in about five days and an estimated 100 million monthly users in about two months, an estimate from a UBS analysis at the time. Those figures, and the way they were measured, are not directly comparable with app installs, but they set the bar.
Six point six million installs in a month would be a hit for most apps. But Muse launches into a market where rivals already have hundreds of millions of users. Alphabet’s Gemini app had 950 million monthly active users, according to figures cited by 24/7 Wall St. Meta’s advantage is distribution: its Meta AI assistant already sits inside WhatsApp, Instagram and Facebook, and Meta reported a 60% jump in daily Meta AI interactions after switching to Muse Spark. If Wonder flows into those apps, Muse’s reach could grow far faster than a standalone app.
Those spending figures explain why Meta needs Muse to work. A company spending well over $100 billion a year on data centres, including the Prometheus site in Ohio where Wonder was trained, needs AI to earn money directly, not only to improve advertising. Paid tiers of $20 and $100 a month are the first test of whether people will pay Meta for an assistant.
Fourteen Months From Reset to Frontier
- Nov 2022OpenAI releases ChatGPT; it reaches an estimated 100 million users in about two months.
- Aug 2025Meta restructures its AI work under Alexandr Wang, the Scale AI founder, as chief AI officer.
- 8 Apr 2026Meta Superintelligence Labs announces its first model, Muse Spark.
- 8 Sep 2026Muse agent app launches on iOS and Android with free and paid tiers; Meta stock rises about 6% the next day.
- 7 Oct 2026Muse arrives on iPad; installs pass 6.6 million.
- 12 Oct 2026Muse Wonder and Wonder Full released through the API and Muse Code.
The speed is part of the story. After a difficult period for its earlier open-weight Llama models, Meta rebuilt its AI organisation and, in about 14 months, produced a model it says competes with the best. Whatever the final verdict on the “biggest launch” claim, that turnaround is the most notable fact here.
Agents, Memory and the Safety Questions
Wonder’s two headline features are aimed at agents. Active Memory carries relevant information across conversations and recalls it when it matters. Active Focus keeps working when a task gets hard, tries other approaches, recovers from errors and checks in before important actions. Research scientist Kavya Kopparapu said in Meta’s launch video: “We put a tremendous amount of effort into making sure your model can remember what’s important to you.”
Memory and the ability to act are also where risk sits. An agent with access to email, calendars and payment accounts can do real damage if it misunderstands an instruction or is tricked by a malicious web page. Meta says Wonder is trained not to share private information without permission and to check in before important actions. Its model card reports refusal rates of 98.7% on a biology misuse test and 0% on an agentic misalignment test, and says all risk areas were rated “moderate or lower” after safeguards.
Meta says a fuller safety report “will be available soon.” Until it is, outsiders cannot check how these tests were run. For anyone connecting bank, shopping or work accounts to an AI agent, the sensible approach is the same whatever the model: start with low-stakes connectors and keep approval for payments switched on.
Is It the Biggest Launch Since ChatGPT? A Scorecard
A launch can be “big” in four different ways. The rail scores Muse on each, based on the evidence available on 14 October 2026.
On capability and price, Muse belongs in the conversation with the strongest launches of 2026. On adoption and independent proof, it is too early to say, and the early adoption data falls well short of ChatGPT’s. “The biggest launch since ChatGPT” is best read as an opinion about potential, not a description of what has happened.
Muse Terms Decoded
| Term | What it means | Figure | Why it matters |
|---|---|---|---|
| Frontier model | A model at the leading edge of capability | Wonder, released 12 Oct 2026 | Meta’s first claimed entry to this tier |
| Muse Wonder Full | Larger version, more compute per answer | Muse Code $50 and $200 plans only | Not on the API |
| Context window | How much text the model can consider at once | 1 million tokens | Long documents and codebases fit |
| Token | A chunk of text the model reads or writes | $4 per million in, $20 per million out | Sets the API bill |
| Cached input | Repeated text billed at a discount | $0.20 per million tokens | Cuts costs for repeated prompts |
| Connector | A linked account the agent can use | Over 10 partners listed | More access, more risk |
| Active Memory | Recall of useful details across chats | Leads MemoryBench, per Meta | Core of a personal assistant |
| Contributor mode | Muse Code setting allowing training on your data | On by default | Switch off for private code |
Should You Try Muse Now?
- Developers: try Wonder on the $5 Muse Code plan or the API with a real task from your backlog, and compare it with your current tool before switching.
- Everyday users: the free Muse tier is enough to judge the agent; Wonder reaches subscribers first, so there is no need to pay just to wait.
- Turn off Contributor mode in Muse Code if you work with private or client code.
- Add connectors slowly. Start with calendar or notes before email, shopping or payments.
- Check availability. Regional pricing, including for India, had not been published at launch.
- Test on 3 to 5 real tasks, not demo prompts
- Compare with your current assistant side by side
- Keep payment approval switched on for agent purchases
- Review which of 10 or more connectors are linked
- Estimate API cost with output at $20 per million tokens
- Recheck independent leaderboards after 4 to 8 weeks
Frequently Asked Questions
What is Meta Muse Wonder?
Muse Wonder is a large AI model released by Meta Superintelligence Labs on 12 October 2026. Meta describes it as its first frontier model, designed for coding and agent tasks, with a 1 million token context window and features called Active Memory and Active Focus.
Is Muse better than ChatGPT?
On Meta’s published coding tests, Muse Wonder scores above GPT-6 Astra, for example 78.8 against 74.1 on DeepSWE 1.1. Those are Meta’s figures and cover coding only. Independent comparisons on everyday tasks were not available at launch.
Is Meta Muse free?
The Muse app has a free tier, with paid plans at $20 and $100 a month. Muse Code, the coding agent, has separate plans from $5 to $200 a month. Wonder reaches Muse app subscribers first.
How much does the Muse Wonder API cost?
$4 per million input tokens, $0.20 per million cached input tokens and $20 per million output tokens. Muse Wonder Full is not available through the API.
What is the difference between Muse Spark and Muse Wonder?
Muse Spark, announced in April 2026, is a lighter, faster model. Muse Wonder is larger and scores higher on complex tasks such as coding. Wonder Full is a further step up, available only on higher Muse Code plans.
Is Meta Muse available in India?
Meta has said availability is subject to region and had not published India pricing or a confirmed India rollout for Wonder at launch. Check the Muse app store listing or Meta’s site for your country.
How many people use Meta Muse?
Sensor Tower estimated more than 6.6 million installs of the Muse app by early October 2026, about a month after launch. Meta has not published active user or subscriber numbers for Muse.
Is it safe to connect my accounts to Muse?
Meta says Wonder is trained not to share private information without permission and to check before important actions. Its full safety report was still to come at launch. Start with low-risk connectors and keep approval for purchases switched on.
The Short Version
Meta’s Muse Wonder, released on 12 October 2026, is the company’s first frontier AI model, and on Meta’s own coding tests it edges ahead of leading rivals. It is priced to compete, from a $5 coding plan to $4 and $20 per million API tokens. But the scores are Meta’s, consumer access is still rolling out, and the Muse app’s 6.6 million installs in a month are far behind ChatGPT’s 2022 pace. It is Meta’s biggest AI launch. Whether it is the biggest since ChatGPT will be decided by users, not launch charts.
