Meta’s AI agent Muse has crossed 3.4 million downloads less than three weeks after launch, and the model powering it just got promoted. Muse Spark 1.3, the latest version of Meta’s in-house model family, has entered what analysts are calling the “frontier tier” of AI models, according to third-party test results cited by J.P. Morgan analysts and reported by Moomoo. The combination of a viral consumer app and a genuinely competitive underlying model is forcing a reassessment of Meta’s place in the AI race, just seven weeks after the company was still widely described as an also-ran behind OpenAI, Anthropic, and Google.
The numbers moving markets this week are download counts, daily active users, and a stock chart that jumped 11% in a single session. But the more consequential shift, according to people tracking the model itself, is technical: Muse Spark 1.3 marks the point where Meta’s models stopped trailing the leaders on coding and agentic benchmarks and started matching them, at a fraction of the price. That combination, a hit consumer app plus a cost-competitive frontier model, is why this week’s news cycle around Meta Muse Spark 1.3 is being treated as more than routine product news.
Muse Spark 1.3 Crosses Into the Frontier Tier
Meta shipped Muse Spark 1.1 in July 2026, describing it at the time as a strong agentic and coding model built for the rebuilt Meta AI assistant. Within roughly two months, the company iterated again, releasing Muse Spark 1.3 as the model layer behind its new Muse consumer agent. According to third-party evaluation results cited in J.P. Morgan’s coverage of the launch, Muse Spark 1.3 has entered the frontier model tier, a marker that puts Meta in the same conversation as OpenAI and Anthropic on raw model capability rather than just app-store velocity.
Meta CEO Mark Zuckerberg framed the release in similarly pointed terms. “Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter,” he said, according to VentureBeat’s coverage of the launch. He also singled out the model’s coding gains specifically: “This is the biggest jump we’ve made so far on coding and agentic work,” Zuckerberg said, per the same report.
Pricing for the new model reflects Meta’s stated strategy of undercutting rivals on cost while closing the capability gap. Muse Spark 1.3 Standard is priced at $1.25 per million input tokens and $4.25 per million output tokens, while a lower-cost contributor version runs $0.10 per million input tokens and $0.20 per million output tokens. That places Muse Spark 1.3 well below the per-token pricing of several rival frontier models, part of why Meta has leaned so heavily on the “too cheap to meter” framing in its own messaging.
Muse Spark 1.3 is not Meta’s first attempt at a distributable model this year. In August, Meta released Muse Glimmer 30B, an open-weight language model with 29.6 billion parameters under an Apache 2.0 license, built to run locally on a single consumer GPU. That release targeted developers who wanted an open, self-hostable model. Muse Spark 1.3, by contrast, is the closed frontier model powering Meta’s flagship consumer product, and it’s the one carrying this week’s headlines.
What “Frontier Tier” Actually Means Here
“Frontier” is not a formally standardized label, and Meta has not published a single, independently audited benchmark suite proving Muse Spark 1.3 beats specific named competitors across the board. What the reporting establishes is narrower but still notable: third-party test results reviewed by J.P. Morgan analysts were strong enough that the firm characterized the release as Meta’s shift “from a laggard to a major competitor,” according to Moomoo’s translation of the analyst note. J.P. Morgan followed that assessment by raising its Meta price target, citing Muse, the Muse Spark model API, and subscription potential as the drivers.
That matters because Meta has spent the better part of two years being characterized by Wall Street as an AI spender rather than an AI builder, pouring capital into data centers and GPUs without a clear consumer product to show for it. Muse Spark 1.3 and the Muse app are the first products where analysts are explicitly crediting Meta’s model quality, not just its infrastructure spend, as the reason to raise price targets.
Muse App Downloads: How Fast Is the Surge, Really?
Meta launched Muse, its personal AI agent, on September 8, 2026. Unlike a standard chatbot, Muse is built to complete multi-step tasks on a user’s behalf, such as filling out forms, booking appointments, logging meals, and sending emails, by connecting to a user’s calendars, payments, and other apps. The rollout has been gradual and US-only so far, available via iOS, Android, and the web.
The download numbers since then have been reported by multiple market-intelligence firms, with some variance between them, which is worth flagging upfront since none of these are Meta’s own official disclosures. Sensor Tower data, cited by TechCrunch, put Muse at 730,000 downloads in roughly its first five days, then 2.5 million downloads by September 21, and finally more than 3.4 million downloads as of a September 25 update reported by TechCrunch. Sensor Tower separately said Muse averaged roughly 55% day-over-day download growth between September 8 and September 17.
Other firms produced different totals, a reminder that these are estimates rather than confirmed figures. Apptopia reported roughly 4.3 million total downloads for Muse, while Appfigures estimated approximately 2.3 million installs as of September 24. On a strictly comparable basis, Apptopia measured 1.8 million iOS downloads for Muse across the US and Canada in its first 12 days, compared with 1.3 million for ChatGPT over its own first 12 days after launch, a gap of roughly 500,000 downloads, or about 38%.
Muse also had a clean crossover moment: it overtook ChatGPT as the top free app on the US iOS App Store on September 18, according to Sensor Tower data cited by CNBC. Three days later, on September 19, Muse logged approximately 264,000 US downloads, its third consecutive day above 200,000 daily downloads.
Muse vs ChatGPT: Comparing Early Adoption Curves
The most-cited comparison in this week’s coverage pits Muse’s first 12 days against ChatGPT’s own early launch numbers, since ChatGPT set the template for viral AI-app adoption when it launched in November 2022. The table below lays out the figures as reported across Sensor Tower, Apptopia, and TechCrunch’s aggregation of that data.
| Metric (first 12 days) | Muse | ChatGPT (comparable window) | Source |
|---|---|---|---|
| iOS downloads, US & Canada | 1.8 million | 1.3 million | Apptopia, via Yahoo Finance |
| Total downloads, US & Canada | 2.8 million | Not disclosed in this comparison | Sensor Tower, via TechCrunch |
| US daily active users | 642,000 | 231,000 | Sensor Tower, via TechCrunch |
| Downloads in first 5 days | 730,000 | Not disclosed in this comparison | Sensor Tower, via TechCrunch |
| Day-over-day growth (Sept 8-17) | ~55% average | Not disclosed in this comparison | Sensor Tower, via TechCrunch |
Downloads alone don’t prove engagement, which is why the daily-active-user comparison matters more than the raw install count. Muse’s 642,000 US daily active users against ChatGPT’s 231,000 over a comparable early window suggests people aren’t just downloading Muse out of curiosity and abandoning it, they’re opening it repeatedly in the first two weeks. That is the pattern investors have been watching most closely, since download spikes are common for any high-profile app launch but sustained daily usage is much harder to manufacture.
Daily Active Users Keep Climbing
Sensor Tower estimated Muse had reached about 560,000 daily active users after 11 days. A separate estimate from measurement firm Similarweb, cited in later coverage, placed Muse near 700,000 US daily users as the app approached the end of September. Muse also reportedly pulled in more downloads during its first two weeks than the mobile apps for Anthropic’s Claude and xAI’s Grok, according to CNBC reporting cited by TheStreet, positioning Muse as the fastest-growing consumer AI app since ChatGPT’s original 2022 launch.
Meta’s own Q2 2026 earnings call offered a preview of this trajectory before Muse even existed as a standalone app. On that call, the company said it had rebuilt Meta AI and integrated Muse Spark into the assistant, and that the change alone had produced a 60% increase in the number of people interacting with the assistant daily. Muse, released roughly six weeks after that call, appears to have accelerated that trend rather than started it.
Wall Street Reacts: Stock Moves and Price Targets
Meta shares rose approximately 11% on Monday, September 21, as investors digested the early adoption data on Muse. Secondary reporting put the single-day gain at roughly $192 billion in added market value, though that figure comes from aggregated market-cap reporting rather than a Meta disclosure. J.P. Morgan followed with a price-target increase to $820, explicitly citing Muse’s download and engagement numbers, the Muse Spark model API, and the subscription tiers as the basis for a more bullish view on Meta’s monetization path. Other analysts, including at Piper Sandler, also revisited their Meta price targets in the days following the launch, according to TheStreet’s coverage of the stock’s reaction.
None of this happened in a vacuum. This week’s Muse coverage lands alongside Meta’s broader AI stock rally, which our earlier report on the Nasdaq’s Meta-driven weekly gain tracked as it rippled through the wider index. It also follows months of scrutiny over Meta’s AI capital spending, the kind of spending we detailed in our coverage comparing Meta’s and Google’s stock performance earlier this year.
Pricing and Monetization: How Meta Plans to Cash In
Muse’s consumer pricing follows a familiar freemium structure, but with an unusual monetization layer bolted on. Free users get 100 million tokens per week, more than enough for casual task delegation. Paid tiers step up to $20 per month for 500 million tokens and $100 per month for 3 billion tokens, aimed at power users who want Muse running errands continuously across email, calendars, and shopping.
The more novel piece is transactional. Meta plans to take a small cut of purchases and bookings that Muse completes on a user’s behalf, with that fee charged to merchants rather than to the user directly. It’s a model closer to how payment processors or travel-booking affiliates monetize than how a typical AI subscription works, and it’s central to why analysts see Muse as a potential new revenue line rather than just another chatbot subscription.
| Tier | Price | Token allowance | Notes |
|---|---|---|---|
| Free | $0 | 100 million tokens/week | US-only rollout, iOS/Android/web |
| Paid, standard | $20/month | 500 million tokens | Consumer tier |
| Paid, power user | $100/month | 3 billion tokens | Consumer tier |
| Muse Spark 1.3 Standard API | $1.25/$4.25 per 1M tokens | Input/output, respectively | Developer access |
| Muse Spark 1.3 Contributor API | $0.10/$0.20 per 1M tokens | Input/output, respectively | Discounted developer tier |
That two-track pricing, cheap consumer subscriptions plus merchant transaction fees plus a discounted developer API, is a broader bet than the one-price-per-seat approach most AI labs have used. It also lines up with the aggressive API pricing Meta used for the earlier Muse Spark 1.1 release, which our prior coverage of Muse Spark 1.1’s debut found was priced as much as 83% below Claude at the time.
Amazon Blocks Muse While Shopify Opens the Door
Muse’s ability to complete purchases on a user’s behalf has already triggered a split response from major retailers. Amazon began blocking the Muse AI agent from its platform on the night of Sunday, September 20, 2026. A day later, Shopify announced it would open every store on its platform to the same agent, taking the opposite approach. The contrast highlights an emerging fault line in e-commerce: platforms that see agentic shopping as a threat to their own checkout funnel versus platforms that see it as a new sales channel worth courting early.
This isn’t the first time a major platform has pushed back against Meta’s Muse agent. Our earlier report on Amazon blocking Meta’s Muse agent covered the initial standoff in more detail, and the dynamic has only sharpened as Muse’s download numbers have grown. Expect more retailers to pick a side in the coming weeks, since the transaction-fee model Meta is using only works if enough merchants are willing to let Muse complete a purchase without a user ever opening their own app or site.
The Muse Spark Model Family: From 1.1 to 1.3
Muse Spark’s version history helps explain how Meta got here so quickly. The family started as the model layer behind a rebuilt Meta AI assistant, with Meta reporting a 60% jump in daily assistant interactions after that rebuild during its Q2 2026 earnings call. Muse Spark 1.1 shipped in July 2026, described internally as a strong agentic and coding model. Meta then iterated again within about two months, arriving at Muse Spark 1.3 in time to power the standalone Muse consumer agent at its September 8 launch. In between, Meta also shipped Muse Glimmer 30B in August as an open-weight, locally runnable alternative for developers who don’t want to depend on Meta’s hosted API.
| Release | Date | Type | Key detail |
|---|---|---|---|
| Muse Spark integration into Meta AI | Reported on Q2 2026 earnings call | Assistant upgrade | 60% increase in daily assistant interactions |
| Muse Spark 1.1 | July 2026 | Closed model | Positioned as a strong agentic/coding model |
| Muse Glimmer 30B | August 10, 2026 | Open-weight (Apache 2.0) | 29.6B parameters, runs on one consumer GPU |
| Muse (consumer agent) | September 8, 2026 | Consumer app | Powered by Muse Spark, US-only rollout |
| Muse Spark 1.3 | ~September 2026 | Closed model | Entered frontier tier per third-party tests |
That cadence, three model iterations plus one open-weight release inside roughly three months, is fast even by 2026 AI industry standards, and it’s a large part of why analysts are revising their view of Meta’s model-building capability rather than just its product-marketing capability.
Competitive Landscape: Where Muse Spark 1.3 Sits
Meta has not published a head-to-head benchmark table pitting Muse Spark 1.3 against specific named rival models, so any direct scoring comparison should be treated cautiously until Meta or an independent lab publishes one. What’s clear from the reporting so far is directional: J.P. Morgan’s note framed Muse Spark 1.3 as marking Meta’s shift from a trailing player to a genuine competitor on agentic and coding tasks, the same categories where OpenAI’s and Anthropic’s flagship models have set the pace over the past year.
Readers looking for a fuller, benchmark-level breakdown of how Meta’s earlier Muse Spark release stacked up against rival frontier models can see our dedicated comparison of Muse Spark against GPT-6 Astra and Claude Fable 5.1, which tracked coding-benchmark performance before this week’s 1.3 upgrade. What has changed since that comparison is pricing and, according to this week’s reporting, model tier: Muse Spark 1.3’s per-token pricing undercuts most rival frontier models by a wide margin, which is the piece of the story Meta has pushed hardest in its own messaging.
Historical Context: How This Compares to Past AI App Launches
ChatGPT’s November 2022 launch remains the reference point every AI app gets measured against, and for good reason: it went from zero to a defining consumer product faster than almost any software release in history. Muse’s early numbers, more downloads and more daily active users than ChatGPT logged in a comparable early window, are the first time a rival consumer AI app has beaten that specific yardstick on Sensor Tower and Apptopia’s own measurements.
It’s worth noting the market Muse is launching into looks nothing like the one ChatGPT launched into. In late 2022, ChatGPT had essentially no direct competition. Muse is launching against an established field that includes ChatGPT itself, Claude, Gemini, and Grok, all of which already have tens of millions of existing users and brand recognition. Beating ChatGPT’s early download curve in that more crowded environment is arguably a stronger signal than the raw numbers alone suggest, since Muse had to pull users away from incumbents rather than simply capture a novel category.
What Analysts and Data Firms Are Saying
Beyond Zuckerberg’s own comments, the numbers behind this week’s coverage come primarily from third-party measurement firms rather than Meta’s own disclosures, and it’s worth hearing how those firms characterized the data directly. Sensor Tower’s own figures, cited in Yahoo Finance’s reporting on Muse’s market impact, state that “Muse reached 2.8 million downloads in its first 12 days across both the US and Canada app stores.” Apptopia’s competing measurement, cited in CNBC’s comparison of Muse against other AI apps, put it more simply: “Muse has clocked over 2.5 million downloads since its debut.”
The gap between those two figures, both measuring roughly the same window, is a useful reminder that app-store download counts are estimates built from sampled data, not exact numbers pulled from Apple’s or Google’s own ledgers. Even so, every firm tracking Muse agrees on the direction: downloads accelerating, daily active users climbing, and Muse holding the top spot on both the Apple App Store and Google Play at multiple points since launch.
What This Means for Developers and Enterprises
For developers, the immediate takeaway is pricing pressure. Muse Spark 1.3’s Standard API pricing, $1.25 per million input tokens and $4.25 per million output tokens, sits well below what several rival frontier-class models charge, and the deeply discounted Contributor tier at $0.10/$0.20 per million tokens gives Meta a plausible path to winning developers who are price-sensitive but still want frontier-tier output quality. Combined with the earlier Muse Spark 1.1 pricing, which our coverage found ran as much as 83% below Claude, Meta’s API strategy looks less like a one-time promotional discount and more like a sustained undercutting campaign.
For enterprises evaluating AI vendors, the more immediate signal is agentic reliability. Zuckerberg’s own framing, that 1.3 represents “the biggest jump we’ve made so far on coding and agentic work,” according to VentureBeat, points directly at the workflows enterprise buyers care about most: multi-step task completion without constant human correction. Whether Muse Spark 1.3 actually delivers on that at enterprise scale, rather than just in consumer app demos, is the open question analysts will be watching for in Meta’s next earnings call.
Predictions: Where Muse and Muse Spark Go From Here
Based on the trajectory reported this week, several developments look likely in the coming months, though these are forward-looking assessments rather than confirmed plans.
- International expansion. Muse’s US-only rollout is the most obvious constraint on its growth curve. A broader international launch would be the clearest way to sustain the download momentum once US market saturation slows things down.
- More retailer standoffs. With Amazon blocking Muse and Shopify embracing it, expect other major retailers to publicly declare a position on agentic shopping traffic in the next quarter, especially as Meta’s merchant-fee model starts generating real transaction volume.
- A Muse Spark 1.4 or similar iteration. Given the pace from 1.1 to 1.3 in roughly two months, another model update before year-end looks plausible if Meta wants to keep pace with expected releases from OpenAI and Anthropic.
- Competitive pricing responses. Muse Spark 1.3’s aggressive per-token pricing is likely to pressure rival labs into further API price cuts, continuing a pattern already visible across the model market through 2026.
- Increased regulatory attention. An AI agent that can independently complete purchases and bookings on a user’s behalf, at this download scale, is likely to draw scrutiny from consumer-protection regulators, particularly around consent and spending controls.
The Bigger Picture for Meta’s AI Strategy
Muse and Muse Spark 1.3 arrive at a moment when Meta has faced sustained criticism over the return on its AI capital spending. A hit consumer app paired with a genuinely competitive model changes that conversation, at least for now, and it’s the first time in this AI cycle that Meta has been credited with product execution rather than just infrastructure scale. Whether that holds depends on whether Muse’s engagement numbers stay elevated once the initial novelty fades, and whether the merchant-fee monetization model actually produces meaningful revenue once more retailers pick a side.
For now, the combination of a chart-topping app and a model upgrade analysts are calling frontier-tier is enough to have moved Meta’s stock, moved at least one major price target, and forced competitors to take notice. The next real test comes when Meta reports how much of that download surge converts into paying subscribers and merchant transaction volume, numbers that, unlike download counts, Meta will eventually have to disclose itself.
Frequently Asked Questions
What is Muse Spark 1.3?
Muse Spark 1.3 is the latest version of Meta’s in-house AI model family, released in September 2026 to power the company’s Muse consumer AI agent. Third-party test results cited by J.P. Morgan analysts placed it in what the firm described as the frontier model tier.
When did Meta launch the Muse app?
Meta launched Muse, its personal AI agent, on September 8, 2026, initially available in the US through iOS, Android, and the web.
How many downloads has Muse had so far?
Estimates vary by measurement firm. Sensor Tower put Muse’s downloads at more than 3.4 million as of a September 25 update, while Apptopia estimated roughly 4.3 million and Appfigures estimated approximately 2.3 million installs as of September 24. All figures are third-party estimates rather than numbers Meta has officially disclosed.
How does Muse compare to ChatGPT’s early adoption?
On a comparable 12-day basis, Muse recorded 1.8 million iOS downloads in the US and Canada versus 1.3 million for ChatGPT, according to Apptopia data. Muse also reported 642,000 US daily active users in its first 12 days versus 231,000 for ChatGPT over a comparable period, according to Sensor Tower data cited by TechCrunch.
How much does Muse cost?
Muse’s free tier includes 100 million tokens per week. Paid tiers cost $20 per month for 500 million tokens and $100 per month for 3 billion tokens. Meta also plans to charge merchants a small fee on transactions Muse completes on a user’s behalf.
Why did Amazon block Muse?
Amazon began blocking the Muse AI agent from its platform on September 20, 2026, according to Bloomberg reporting cited by TheStreet. Amazon has not publicly detailed its reasoning, but the move came as Muse’s ability to complete purchases on a user’s behalf raised questions about how agentic shopping tools interact with existing retail platforms. Shopify took the opposite approach, opening its platform to Muse the following day.
What is Muse Glimmer 30B?
Muse Glimmer 30B is a separate, open-weight language model from Meta with 29.6 billion parameters, released under an Apache 2.0 license on August 10, 2026. Unlike Muse Spark, which powers Meta’s hosted Muse app, Muse Glimmer is designed to run locally on a single consumer GPU.
Is Muse available outside the United States?
As of this week’s reporting, Muse’s rollout remains US-only. Meta has not announced a confirmed international launch date.