Can Meta’s Muse catch ChatGPT?

Last updated: 21 October 2026
Cover of NewMarketPitch's Frontier AI Labs: Who's Ahead report ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.What is Meta spending on its AI unit, and what has it shipped? About $140B of spending planned in 2026 for all of Meta. Its new lab has shipped four Muse products, used by just over 3 million people a week. Chart from NewMarketPitch's Frontier AI Labs report.Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.Who is fighting on which front of frontier AI? Chat assistants are the most crowded front, with big tech, the US labs and three Chinese rivals. Europe shows up mostly through Mistral, on three fronts. Chart from NewMarketPitch's Frontier AI Labs report.Which lab makes the best open-weight models? Moonshot AI on test scores, with Kimi at 46 against 32 for DeepSeek and Zhipu. Alibaba's Qwen is downloaded the most. Chart from NewMarketPitch's Frontier AI Labs report.Who is really ahead in frontier AI, OpenAI or Anthropic? Anthropic, on 3 lines of 5: $17B of third-quarter sales against about $10B, and the top model on the experts' index. OpenAI keeps the crowd and has raised more money. Chart from NewMarketPitch's Frontier AI Labs report.How much electricity will frontier AI labs need? As much as whole countries. OpenAI aims for about 30 gigawatts, close to what Spain uses on average, and Anthropic has deals for about 15, more than Belgium. Chart from NewMarketPitch's Frontier AI Labs report.

Every funding round, every investor, and what the labs promised in return, in one document. Have a look ↓

Summary

Meta's Muse can catch ChatGPT in AI model quality and could eventually lead in personal AI assistance, but overtaking ChatGPT in overall users and commercial adoption remains a distant prospect. Meta has closed the technology gap much faster than the adoption gap.

Muse Wonder Full has reached frontier-model performance in just months. It scored 56 on the Artificial Analysis Intelligence Index against 53 for GPT-6 Astra and outperformed OpenAI's model on several coding and professional benchmarks, although the strongest Wonder model is not yet available to ordinary Muse users. Three points on one index is a thin lead, so we check which labs really lead on benchmarks and which only claim it in our frontier AI labs report.

Muse's mobile launch was exceptional. It reached 10 million US downloads in 40 days, compared with 123 days for ChatGPT, and became the fastest AI assistant to reach that milestone in Sensor Tower's historical comparison. But 94% of Muse downloads came from the United States, showing how geographically concentrated the launch remains.

Early retention tells a more complicated story than download rankings. Muse retained 30% of new users after seven days and 23% after fourteen, outperforming comparable AI apps, but its day-one retention of 73% fell by 50 percentage points within two weeks. The initial excitement is fading faster than the headline comparisons suggest.

Meta's advertising campaign is a major part of the equation. Muse accounted for an estimated 39% of Meta's advertising spending in October, and a significant promotional increase coincided with a 73% jump in daily downloads. Meta has demonstrated its ability to manufacture awareness at extraordinary scale, but organic demand remains harder to measure.

The absolute audience gap is still enormous. ChatGPT reported more than one billion weekly users in September 2026, while early Muse disclosures indicated more than four million. Even allowing for Muse's subsequent growth to 2.5 million US daily mobile users, Meta remains far behind OpenAI's established audience.

Meta's most interesting advantage may be distribution rather than model intelligence. WhatsApp, Instagram, Facebook, and eventually smart glasses give Muse access to billions of existing users. Even a small share adopting Muse regularly could turn it into a major personal-agent platform without requiring everyone to download another app.

Muse is also competing in a more ambitious category than traditional chatbots. It can browse websites, schedule activities, send approved emails, and make purchases. Yet Amazon's decision to block Muse purchases shows that autonomous agents depend on third-party permissions as much as technical capability.

ChatGPT retains a substantial advantage in professional adoption, paid subscriptions, and established workflows. Meta can afford to subsidize Muse through its advertising business, but it has not disclosed enough subscription revenue or operating-cost data to demonstrate sustainable economics for autonomous agents.

Our conclusion is that Muse is already a credible competitor and may become particularly strong at everyday personal assistance. The real test is whether users continue returning after Meta reduces its promotional pressure, trust Muse with sensitive tasks, and find its integrations genuinely more convenient than ChatGPT. Model benchmarks have made the competition close; habitual usage will decide whether Meta can close the much larger commercial gap.

Written by the New Market Pitch teamWe track new markets so founders and investors can move faster. About us

ChatGPT, Gemini, Meta AI, Claude: who has the most users? ChatGPT, with 1.2 billion people a week. Gemini and Meta AI each claim about a billion a month, and Anthropic gives no count for Claude. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

Why is Meta's Muse suddenly challenging ChatGPT?

Meta's Muse has become a serious ChatGPT challenger remarkably quickly, thanks to a new generation of AI models and one of the fastest mobile-app launches in recent years.

Meta introduced Muse Spark in April 2026, followed by the standalone Muse personal agent in September and the more advanced Muse Wonder models in October. Within roughly six months, Meta went from rebuilding its AI technology to releasing models capable of competing with OpenAI's strongest systems on difficult reasoning and coding tasks.

Consumer adoption followed. According to Sensor Tower's October 20 report, Muse reached 10.9 million mobile downloads in approximately six weeks, including 10.2 million in the United States. The app also reached 2.5 million US daily active users on October 16. Ten million downloads in six weeks is a fast start; we go through how much has moved in a year in our report on frontier AI labs.

The competition has changed for another reason: Muse is designed to complete tasks on people's behalf. It can browse websites, manage schedules, send approved emails, and make purchases. Meta wants people to delegate everyday work to an assistant that remembers their preferences.

ChatGPT has already developed similar agent capabilities, but its position is much more established. OpenAI reported more than one billion weekly ChatGPT users in September 2026.

Muse has entered the market with competitive technology and millions of users. Whether those users keep coming back will tell us much more than its impressive launch.

What would it actually take for Muse to catch ChatGPT?

Muse could catch ChatGPT in personal AI assistance long before it matches ChatGPT's overall user base. The two products increasingly perform similar tasks, but their strongest use cases and established audiences are different.

ChatGPT already serves people researching topics, writing documents, programming, analyzing files, and managing professional work. OpenAI has also expanded into autonomous tasks through ChatGPT Work and Dots.

Muse is concentrating on personal assistance, particularly tasks involving schedules, messaging, shopping, and information spread across connected applications.

That difference gives Meta a realistic opportunity. Someone might continue using ChatGPT for research or coding while choosing Muse to arrange appointments, organize purchases, or manage daily responsibilities.

For Muse to catch ChatGPT overall, however, we would need evidence that millions of people use it regularly, trust it with important tasks, and increasingly prefer it over competing assistants. Matching OpenAI's model performance would only solve one part of that challenge.

Is Meta's Muse Wonder as smart as ChatGPT now?

Meta's Muse Wonder Full is already competitive with GPT-6 Astra, and some of its benchmark results are better. Meta has clearly returned to the frontier of AI development.

In evaluations published around Meta's October 12 launch, Muse Wonder Full scored 56 on the Artificial Analysis Intelligence Index, compared with 53 for GPT-6 Astra and 58 for Claude Opus 5.5.

The results were particularly strong in software engineering. Wonder Full scored 79.0 on DeepSWE 1.1 against Astra's 74.1, and 57.5 on SWE-Marathon against 42.5. It also led on Harvey's legal-work benchmark and JobBench, which evaluates professional tasks.

OpenAI performed better elsewhere. Astra scored 59.1 on Terminal-Bench 4.0-AA versus Wonder Full's 57.1, and 65.3 on HealthBench Pro versus 62.3.

These comparisons come with a qualification: several evaluations were conducted or assembled by Meta, and some used different testing configurations. We have enough evidence to call Muse Wonder a frontier model, although independent testing across everyday tasks is still needed before declaring an overall winner.

There's another distinction. Wonder Full is available to Muse Code subscribers, while the consumer Muse agent has not yet received the full Wonder upgrade. The published benchmark results therefore describe Meta's strongest technology rather than what every Muse user can experience today.

Benchmark Muse Wonder Full GPT-6 Astra Leader
Artificial Analysis Intelligence Index 56 53 Muse
Harvey LAB-AA, legal work 13.1 8.6 Muse
AutomationBench-AA 45.4 41.6 Muse
JobBench, professional tasks 75.3 66.1 Muse
DeepSWE 1.1 79.0 74.1 Muse
SWE-Marathon 57.5 42.5 Muse
Terminal-Bench 4.0-AA 57.1 59.1 ChatGPT model
HealthBench Pro 62.3 65.3 ChatGPT model

The distinction between Muse Wonder and Muse Wonder Full is important here. Artificial Analysis reports a score of 52 for standard Muse Wonder, versus 56 for Wonder Full. The comparison page for standard Wonder therefore shows 52 against Astra's 53. These are different model variants, not conflicting scores for the same model.

What is Meta spending on its AI unit, and what has it shipped? About $140B of spending planned in 2026 for all of Meta. Its new lab has shipped four Muse products, used by just over 3 million people a week. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

How did Meta catch up with OpenAI so quickly?

Meta has rebuilt its AI capabilities at an unusually fast pace, although we cannot establish that its models are improving faster than OpenAI's. The important change is how much more capable Meta's models have become within the same development cycle.

Meta started with Muse Spark in April 2026. Spark 1.1 followed in July, Spark 1.2 and Muse Code arrived in August, and Spark 1.3 appeared in September. October brought Muse Wonder and Wonder Full.

The company also introduced image and video generation models, Muse Glimmer for smaller-scale agent applications, and voice-related capabilities. This was a much broader expansion than a series of incremental chatbot improvements.

Meta's research organization attributes the progress to an extensive overhaul of its training methods, including pretraining and reinforcement learning. Muse Wonder was trained using the company's Prometheus infrastructure, a gigawatt-scale data-center project.

The speed is especially striking given Meta's earlier difficulties developing successors to Llama that could match the most advanced proprietary models.

OpenAI has continued releasing stronger models and products during the same period, so a direct comparison of improvement rates would require equivalent evaluations, training costs, and deployment data. Those figures are unavailable.

Still, Meta has accomplished something substantial. In six months, it developed a competitive model family covering reasoning, coding, multimodal tasks, and autonomous agents. The next challenge is putting that technology into products people consistently choose.

Is Muse growing faster than ChatGPT did?

Muse is growing much faster than ChatGPT did during its initial US mobile launch, although Meta benefits from entering an AI market that ChatGPT helped create.

Sensor Tower estimated that Muse reached 10 million US downloads in 40 days. ChatGPT needed approximately 123 days to reach the same milestone, while Grok took 211 days and Claude 673 days. Beating ChatGPT's pace to 10 million US downloads changes the picture, and we work out who is ahead in our report on who's ahead in frontier AI labs.

Looking at the same first 40 days, Muse recorded about 10.2 million US downloads compared with ChatGPT's 3.7 million. That makes Muse's initial US installation volume approximately 2.8 times larger.

There's an even more revealing comparison. Sensor Tower has tracked mobile applications since October 2015, and only 11 other apps reached 10 million US downloads within 40 days. Nine were mobile games, one was an entertainment app, and one was Meta's Threads. Muse is the first AI assistant in that group.

However, Muse's growth is heavily concentrated in one market. Approximately 94% of its downloads came from the United States. ChatGPT's initial global app adoption was more geographically diverse, with the US accounting for roughly one-quarter of downloads during its first 40 days.

ChatGPT also launched its mobile app after already attracting users through its website. By the time Muse appeared, consumers were familiar with AI assistants and had clear expectations about their usefulness.

So Muse's launch is exceptional, particularly for an AI product, but the historical comparison doesn't establish that Meta will eventually have more users than OpenAI.

Are people actually using Muse after downloading it?

Muse's early retention is better than that of several competing AI apps, but there is a concerning drop in usage after the first day.

Sensor Tower reported that 30% of people who installed Muse in September returned after seven days, and 23% returned after 14 days. The corresponding averages for ChatGPT, Claude, and Grok were 19% after seven days and approximately 16–17% after 14 days.

That leaves Muse around six or seven percentage points ahead of its competitors after two weeks. It is a meaningful advantage for a newly launched product.

But there's another pattern in the same report. Muse's day-one retention was 73%, falling to 23% on day 14. That's a 50-percentage-point decline. The comparison apps experienced an average decline of around 12 points over the same interval.

Muse therefore starts with exceptionally high initial engagement, then loses users much faster. Its early advantage survives, but narrows considerably.

The usage numbers are still encouraging. Sensor Tower estimated that Muse reached 2.5 million daily active US users on October 16, with daily activity growing 15% week over week during October 11–17.

We also need to keep the measurement limits in mind. Sensor Tower uses modeled estimates, its comparison cohorts come from different product launches, and the firm has not yet provided mature 30-day or 90-day retention data.

Muse clearly has millions of genuine users. What we still don't know is whether its strongest users will remain active once the launch excitement and promotional spending ease.

Who has held the top AI model, month by month? Anthropic and Google, taking turns: over the last 12 months Anthropic topped the LMArena users' vote for 7 months and Google for 5. OpenAI never did. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

How far behind ChatGPT is Muse in actual users?

ChatGPT remains vastly larger than Muse, even after Meta's explosive launch. The available figures put OpenAI's established audience in the hundreds-of-millions-to-one-billion range, while Muse has only recently reached millions of regular users.

OpenAI reported more than one billion weekly active ChatGPT users in September 2026. Internal Meta figures reviewed by The Information around Muse's launch indicated more than four million weekly users interacting with Muse, including more than three million who submitted prompts.

Those figures imply a gap of approximately 250 to one, although they refer to different reporting periods and are not a synchronized measurement of market share.

Sensor Tower's later estimate of 2.5 million US daily active Muse users confirms that its audience has grown considerably. Still, we cannot infer its global weekly or monthly audience accurately from that single-day estimate.

ChatGPT also crossed one billion global monthly mobile-app users in May 2026, according to Sensor Tower. Muse's 10.9 million cumulative downloads remain far below that established base.

This scale difference changes how we should interpret growth. Adding another five million weekly users would be a major breakthrough for Muse. For ChatGPT, five million represents less than 1% of its reported weekly audience.

Adoption measure ChatGPT Meta Muse
Consumer product debut November 2022 September 2026
Reported weekly users More than 1 billion More than 4 million in early reporting
Recent US daily mobile users Not established in a comparable disclosure 2.5 million estimated
Global monthly mobile-app users Above 1 billion in May 2026 Not established
First 40 days of US mobile downloads 3.7 million 10.2 million
Time to 10 million US mobile downloads 123 days 40 days
Seven-day retention, comparison cohorts Included in 19% rival average 30%
Fourteen-day retention, comparison cohorts Included in 16–17% rival average 23%
Initial mobile audience International 94% of downloads from US
Disclosed paying subscribers 50 million in February 2026 Not disclosed

Is Meta buying Muse's popularity through advertising?

Meta's enormous advertising campaign has played a major role in Muse's growth, although the app's retention figures suggest that some users are finding genuine value after installing it.

Sensor Tower estimated that Muse accounted for 39% of Meta's advertising spending during October, up from 23% in September. Muse also climbed from 73rd to 21st among US brands ranked by advertising impressions.

That's an extraordinary amount of promotional attention for a product launched only weeks earlier.

The timing offers another clue. According to Sensor Tower's earlier analysis, Meta increased its Muse advertising campaign around September 16, when daily downloads jumped 73% from the previous day. The company also promoted the assistant heavily during Meta Connect.

We can see the effect in Muse's app-store rankings. It occupied the top position on US Google Play for 30 consecutive days through October 19 and spent almost the entire corresponding period at the top of Apple's US App Store.

The question is how much of this activity would have happened without the campaign. Sensor Tower does not provide the spending amounts or a reliable estimate of downloads attributable specifically to advertising.

Meta's own history offers a useful warning. Threads passed four million US downloads on its first day in 2023, substantially faster than Muse's initial launch. Distribution can create explosive interest, but durable engagement develops on a different timeline.

Muse's unusually high day-one retention followed by a steep decline gives us reason to be cautious. Meta has demonstrated that it can attract users at scale. We need longer retention data to judge how much demand exists independently of its advertising effort.

Who is fighting whom in frontier AI? OpenAI leads Google on users, Anthropic leads OpenAI on sales, and Alibaba leads DeepSeek on downloads. Alibaba and Kling split video, SSI and Thinking Machines split money and product. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Can WhatsApp, Instagram and Facebook help Muse overtake ChatGPT?

Meta's social and messaging applications give Muse a distribution opportunity that OpenAI cannot easily reproduce. The company reported 3.60 billion daily active people across its family of applications in its second-quarter 2026 results.

The potential audience is enormous, particularly because Muse can be used through WhatsApp in supported markets.

Messaging is a natural environment for personal assistance. People already use WhatsApp to coordinate appointments, discuss purchases, arrange travel, and communicate with businesses. Muse could help complete some of those tasks without requiring users to open a separate application. Using apps people already open is one way to find an opening, and we weigh where the biggest opportunities are in our frontier AI labs report.

Meta also plans integrations involving smart glasses. At Meta Connect in September, the company outlined additional Muse features including computer use on Mac, new connectors, video interaction with its avatar, and future glasses integration.

The advantage comes with an important practical limit. A user talking to friends on WhatsApp has not necessarily agreed to let an AI agent read conversations, access personal documents, or send messages on their behalf.

Meta needs users to make those connections voluntarily and trust Muse enough to keep using them.

We also don't have a reliable breakdown showing how many Muse users come from WhatsApp versus its standalone app. That makes it impossible to measure how effectively Meta is converting its existing audience.

Still, the opportunity is clear. Even a modest conversion rate across Meta's existing applications could produce tens of millions of Muse users without requiring the company to build a new distribution network from scratch.

Can Muse actually do things ChatGPT cannot?

Muse has made autonomous personal assistance the center of its product, but ChatGPT already offers many of the same capabilities through Work and Dots. Meta's strongest distinction is how closely it wants to connect an agent with everyday messaging and personal activities.

Muse operates through a cloud-based virtual machine with browser access and connected applications. The agent can research products, fill forms, arrange schedules, send approved emails, and continue working while the user is away.

Meta expanded those capabilities in September 2026 through Stripe's Link payment system, which enables supported purchases using one-time-use payment cards. Muse can therefore move from finding a product to buying it, subject to the relevant approvals and merchant support.

OpenAI has developed similar functionality. ChatGPT Work can operate across files and applications, research complex questions, browse websites, and produce completed work. Dots can remember ongoing responsibilities and make progress between conversations.

The biggest difference is the intended experience. Muse wants autonomous assistance to feel like messaging a personal helper. ChatGPT already has a larger ecosystem of research, coding, and professional tools.

There are some early signs that Muse is becoming relevant outside its own app. DataDome reported that Muse accounted for 33% of agentic browser requests on its network during September, despite being measured for only nine days.

That figure covers DataDome's observed network traffic, not the entire internet. But it suggests Muse is already generating significant real-world agent activity.

The same development has created friction. Amazon blocked Muse from making purchases on Amazon.com in September, showing how merchants can limit an agent's ability to complete transactions.

Muse has meaningful functionality today. Its real advantage will depend on whether it can perform these tasks more reliably and conveniently than ChatGPT, particularly when third-party services don't cooperate.

Who is fighting on which front of frontier AI? Chat assistants are the most crowded front, with big tech, the US labs and three Chinese rivals. Europe shows up mostly through Mistral, on three fronts. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is just a preview page. Don’t worry if a number looks old: the report gets refreshed, and you’ll always get the latest version.

Could Muse's memory make it a better personal assistant than ChatGPT?

Muse Wonder has shown stronger results than GPT-6 Astra on Meta's reported memory benchmark, but we don't yet know whether that produces better personal assistance in everyday life.

Meta developed two features called Active Memory and Active Focus. Active Memory retrieves useful information from earlier conversations, while Active Focus helps the model continue working on difficult tasks, recover from obstacles, and pursue longer-term goals.

On Meta's MemoryBench evaluation, Wonder Full scored 85.5 and standard Wonder 85.6, compared with 81.4 for GPT-6 Astra.

That's an advantage of roughly four points, although MemoryBench is an internally developed evaluation. We would need broader independent testing to know how consistently it applies.

The practical ambition is easy to understand. Someone could ask Muse to organize a family trip, and the assistant might remember preferred destinations, available dates, dietary restrictions, and previous bookings without requiring everything to be explained again.

ChatGPT already provides personalized memory, while Dots can maintain responsibilities using remembered context and connected applications. Both companies are pursuing assistants that become more useful as they learn about their users.

For now, Meta's published results suggest an improvement in memory-related model capabilities. The more important test will be whether Muse remembers the right information, respects privacy boundaries, and completes tasks without repeatedly asking users to correct it.

Is Meta's Muse already better than ChatGPT for coding and professional work?

Muse Wonder Full can outperform GPT-6 Astra on selected coding and professional benchmarks, but ChatGPT currently offers a much more established environment for professional customers.

Meta's SWE-Marathon result is particularly interesting. Wonder Full scored 57.5 against Astra's 42.5, a 15-point advantage on an evaluation involving longer-running software tasks. Muse also performed strongly on DeepSWE 1.1 and JobBench. Winning on long-running software tasks shows which technologies are gaining traction, and we follow those in our frontier AI labs market report.

Those results suggest Meta's recent improvements extend well beyond casual chatbot conversations.

However, OpenAI remains highly competitive on other tests, and benchmark scores don't automatically translate into better developer workflows. Reliability across a complete project, integrations with existing tools, and the ability to recover from mistakes all affect whether professionals adopt a product.

OpenAI has a substantial head start here. ChatGPT Work, Codex, enterprise subscriptions, developer APIs, and workplace integrations already support professional usage. OpenAI reported serving more than 2.5 million businesses across its products in September.

Meta has launched Muse Code and expanded the Meta Model API, including the Muse Wonder family. It also released smaller and open-weight models, such as Muse Glimmer and Muse Spark Open, allowing developers to build on its technology in different environments.

Those releases could make Meta's models more attractive to developers who want to control their deployments. But we haven't seen evidence that Muse has approached ChatGPT's enterprise adoption or paid professional usage.

Meta can now challenge OpenAI on individual technical tasks. Catching up commercially in professional AI will take considerably more time.

Can Meta afford to compete with OpenAI for years?

Meta has enough financial resources to sustain a prolonged fight with OpenAI, even if Muse generates little direct revenue initially. The company's existing advertising business gives it an advantage that most standalone AI companies lack.

Meta generated $60.8 billion in revenue and $31.86 billion in operating cash flow during the second quarter of 2026.

The scale of its AI investment is enormous. Meta's guidance later pointed to full-year capital expenditures of $135 billion to $145 billion, covering AI infrastructure alongside other infrastructure needs.

That spending supports projects such as the Prometheus data center, where Meta says Muse Wonder and Wonder Full were trained.

OpenAI has significant financial resources and a growing commercial business of its own. In February 2026, it reported 50 million paying ChatGPT subscribers, demonstrating that users are already willing to pay directly for AI services.

Meta's position is different. It can make Muse available at low prices or free to many consumers while funding the product through cash generated elsewhere in the company.

However, the economics of autonomous agents could become expensive. A single task may involve browsing dozens of pages, calling a model repeatedly, using connected applications, and running a cloud computer for an extended period.

Meta currently charges $4 per million input tokens and $20 per million output tokens for Muse Wonder through its API. Those are premium frontier-model prices, and they don't reveal the full cost of running the consumer agent.

We don't know Muse's subscription conversion rate, revenue, or cost per completed task. But Meta's finances make a near-term funding shortage unlikely. Its bigger challenge is finding enough recurring use cases to justify continued spending.

Which lab makes the best open-weight models? Moonshot AI on test scores, with Kimi at 46 against 32 for DeepSeek and Zhipu. Alibaba's Qwen is downloaded the most. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Just a preview. If a number here looks old, no worries: your report comes with the latest data.

Is Muse safe enough to manage people's emails, money and personal information?

Muse has a serious security architecture, but there isn't enough independent evidence to conclude that it is safer than ChatGPT when handling sensitive tasks.

Meta runs Muse agents inside isolated cloud-based virtual machines. Its security system includes a separate Sentinel agent that monitors activity and controls certain interactions with the internet.

Credentials are kept separate from the model, while actions such as sending emails and making purchases can require explicit approval. Meta also says Muse conversations and virtual-machine data aren't shared with its advertising systems.

These protections address an important problem. An assistant that reads a webpage could encounter malicious instructions telling it to reveal private information, transfer data, or perform an unauthorized action.

OpenAI faces similar risks with ChatGPT Work and Dots. Its agent products also use cloud environments, access controls, permissions, and approvals for sensitive operations.

Neither company's architecture can eliminate every failure. Agents may misunderstand instructions, interact with compromised services, or take incorrect actions despite technical safeguards.

The financial stakes are also becoming clearer as these products enter commerce. When Amazon blocked Muse from purchasing through its website in September, it showed that autonomous agents depend partly on what external services will permit.

We would need independently measured attack rates, unauthorized actions, and successful task completion data to identify a security leader. For now, Muse's design looks credible, but its safety record remains young.

Can Muse make money without becoming another expensive Meta project?

Muse has several ways to generate revenue, but Meta has not yet shown that people will pay enough for the assistant to cover the cost of running it.

Subscriptions are the most obvious route. Muse offers much of its everyday functionality for free while reserving more intensive capabilities for paying customers. Free everyday use with paid upgrades is one route to revenue; we compare which business models are winning in our frontier AI labs market report.

ChatGPT provides a useful comparison. OpenAI reported 50 million paying subscribers in February 2026, alongside approximately 900 million weekly users. That is equivalent to about 5.6% of the weekly audience, although the figures don't represent a formal conversion-rate calculation.

If Muse eventually reached 50 million weekly users and achieved a similar ratio, that would imply around 2.8 million paying subscribers. This is a scale illustration rather than a revenue forecast, because Muse's audience, pricing, and usage patterns differ substantially.

Meta could also benefit indirectly. Muse might help consumers discover products, support business interactions, or increase usage of WhatsApp and other Meta applications.

The company has already extended Muse to small businesses, allowing it to access advertising accounts and a range of business tools. That opens the possibility of commercial usage beyond individual consumers.

However, every additional task can consume computing resources, especially when agents browse websites or perform long sequences of actions.

Meta has the money to subsidize Muse for years, but a successful launch and a sustainable business are very different achievements. We still need disclosed revenue, paid-user growth, and evidence that the cost of serving each user is falling.

Who is really ahead in frontier AI, OpenAI or Anthropic? Anthropic, on 3 lines of 5: $17B of third-quarter sales against about $10B, and the top model on the experts' index. OpenAI keeps the crowd and has raised more money. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

This is a preview page. Spotted an outdated number? That’s normal here. The report you get is always the freshest version.

What could prevent Muse from catching ChatGPT?

Muse's biggest risks are weak long-term retention, unreliable real-world actions, and the possibility that its strongest models never become a clear consumer-product advantage.

The retention problem is already visible. Sensor Tower's September cohort returned at a rate of 73% after one day and 23% after two weeks. Although Muse remains ahead of the comparison apps at day 14, a large share of its early users aren't returning regularly.

Another challenge is the gap between Muse Wonder's capabilities and the standalone product. Meta has announced that Wonder will reach Muse users gradually, starting with subscribers. Until that rollout is complete, the company's best benchmark scores cannot be taken as evidence of the experience available to everyone.

There is also the practical problem of working across third-party applications. Muse can browse and perform transactions, but Amazon's decision to restrict purchasing shows that even capable agents can encounter external barriers.

These challenges become more serious as agents handle sensitive actions. A wrong answer in a chatbot is frustrating; an incorrect purchase, message, or account change can have financial or personal consequences.

Meanwhile, OpenAI continues improving ChatGPT, Codex, Work, and Dots. Meta will have to win users from an increasingly capable product with years of established habits behind it.

We see no evidence that Muse is about to run into a technical or financial wall. The harder question is whether millions of people will find it reliable enough to use every week without needing a major advertising campaign to bring them back.

Can Meta's Muse catch ChatGPT?

Yes, Meta's Muse can catch ChatGPT in AI model quality and could become a major competitor in personal assistance. But overtaking ChatGPT as the world's most widely used general-purpose AI assistant is still a distant prospect.

On technology, the race is already close. Muse Wonder Full competes with GPT-6 Astra on frontier evaluations and leads on several difficult coding and professional benchmarks. Meta has overcome much of the technical gap that previously separated it from the strongest AI developers.

On consumer growth, Muse has achieved something genuinely unusual. Sensor Tower recorded 10.2 million US downloads in 40 days, making it the fastest AI assistant in the firm's historical comparison. Early retention also exceeds that of several established alternatives, even though usage drops sharply after the first day.

Meta has another advantage through Facebook, Instagram, WhatsApp, and its planned smart-glasses integrations. Those products give Muse access to an enormous potential audience.

But ChatGPT currently operates at an entirely different scale, with more than one billion weekly users, tens of millions of paying subscribers, and a much larger established professional ecosystem.

The decisive test for Muse will be what happens after its early promotional campaign. If people keep returning, trust the agent with important tasks, and find it more convenient than opening ChatGPT, Meta could build one of the world's leading personal AI products.

Our judgment is that Meta has caught up much faster on AI technology than on actual product adoption. Muse is already a credible challenger, and personal agents are a market where it has a realistic chance to lead. ChatGPT, however, remains decisively ahead in overall usage and commercial adoption. That lead will take years of sustained growth to overturn, even with Meta's distribution and financial resources.

How much electricity will frontier AI labs need? As much as whole countries. OpenAI aims for about 30 gigawatts, close to what Spain uses on average, and Anthropic has deals for about 15, more than Belgium. Chart from NewMarketPitch's Frontier AI Labs report.
Frontier AI Labs reportThis page and 70+ morePDF Get the report

Preview only. Some numbers may have changed since we made this page, but your report will be up to date.

Our methodology

This analysis examines whether Meta's Muse can catch ChatGPT by comparing their technological capabilities, consumer adoption, retention, autonomous-agent functionality, distribution, professional usage, and financial resources. We focused on recent, measurable developments rather than treating model announcements or download rankings as sufficient evidence of competitive leadership.

For model performance, we compared Muse Wonder and Muse Wonder Full with GPT-6 Astra using Artificial Analysis results and Meta's published technical evaluations. We distinguish the two Wonder variants: Artificial Analysis reports 52 for standard Wonder and 56 for Wonder Full, while GPT-6 Astra scores 53. We also examined coding, professional-work, memory, terminal-use, and health-related benchmarks, recognizing that some were conducted by Meta or used different evaluation configurations.

For adoption, we used Sensor Tower estimates reported by TechCrunch and EMARKETER, including first-40-day downloads, daily active users, geographic concentration, seven-day and fourteen-day retention, and advertising activity. Historical comparisons use the original mobile launch periods of competing assistants. These figures are modeled estimates, and we distinguish installations from active users and retention from long-term commercial adoption.

We assessed ChatGPT's established scale using OpenAI's reported weekly users, paying subscribers, business customers, and professional product capabilities. We compared these figures with Muse's disclosed and estimated audience, while avoiding direct market-share conclusions from different reporting periods or incompatible weekly, daily, and monthly metrics.

For autonomous-agent capabilities, we reviewed Meta's product launches, technical documentation, security architecture, and announced integrations alongside OpenAI's ChatGPT Work and Dots. We also considered DataDome's observed agentic browser traffic and reported merchant restrictions. These sources help distinguish available functionality from future integrations and show where real-world execution depends on third-party services.

For financial sustainability, we used Meta's second-quarter 2026 results, SEC filings, capital-expenditure guidance, and disclosed model API pricing. We treated possible subscription conversion, future monetization through Meta's applications, and agent operating economics as scenarios rather than established revenue forecasts.

Our conclusion weighs these dimensions separately before bringing them together. Technical competitiveness and rapid downloads demonstrate that Muse is a serious new challenger, while returning-user behavior, established audience size, professional adoption, and task reliability provide a stronger basis for judging whether Meta can eventually catch ChatGPT.

Key sources include: Meta's Muse Wonder launch announcement, Meta AI Research's Muse Wonder technical report, Artificial Analysis's Muse Wonder and Wonder Full evaluation, Artificial Analysis's standard Wonder versus GPT-6 Astra comparison, Meta's Muse personal-agent announcement, TechCrunch's reporting on Sensor Tower's launch research, EMARKETER's Muse engagement analysis, OpenAI's disclosure of ChatGPT and business adoption, Meta's Muse Code and Spark development report, Meta Connect 2026 announcements, DataDome's Q3 2026 AI traffic report, Meta's Muse security architecture, OpenAI's ChatGPT Work announcement, OpenAI's Dots documentation, Meta's Q2 2026 financial results, and Meta's SEC quarterly filing.