Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Mark Zuckerberg announced Meta Superintelligence Labs (MSL) on June 30, 2025, as a reorganization of Meta’s AI work—not as the launch of a superintelligent system. Alexandr Wang was named to lead the effort, with Nat Friedman responsible for AI products and applied research. Since then, Meta has introduced the Muse model family and expanded Meta AI’s task-oriented features, but those releases are progress toward the company’s stated ambition, not proof that superintelligence has been achieved.
What Zuckerberg announced
The June 30, 2025 announcement established MSL as an internal organization bringing together Meta’s foundation-model work, AI product teams, its Fundamental AI Research organization (FAIR), and a new group focused on next-generation models. Meta described the effort as spanning research, model training, infrastructure, and products. It was a company restructuring and strategic push, not a standalone consumer app or a claim that Meta had already built superintelligence. Bloomberg reported the announcement; Meta later discussed the organization in Zuckerberg’s Q2 2025 remarks.
The public story unfolded in stages: Meta named additional leadership in July 2025, Zuckerberg published his “personal superintelligence” vision on July 30, 2025, and the first model Meta identified as part of MSL’s new family, Muse Spark, arrived on April 8, 2026. Meta announced Muse Spark 1.1 and additional Meta AI capabilities in July 2026.
Who leads MSL
| Person or group | Role described by Meta or reporting |
|---|---|
| Alexandr Wang | Leads the overall effort and serves as Meta’s chief AI officer. He previously led Scale AI. |
| Nat Friedman | Responsible for AI products and applied research. |
| Shengjia Zhao | Named chief scientist for the new model-development effort in July 2025; Reuters described him as a ChatGPT co-creator and former OpenAI researcher. |
| FAIR and foundation-model teams | Existing Meta organizations included in the broader MSL structure; they are not simply one lab personally run by Wang. |
Meta outlined Wang’s and Friedman’s roles in its Q2 2025 remarks. Reuters covered Zhao’s appointment in its report on the chief scientist role.
Recommended Free Tools
#1 Best Overall
Why Meta created the lab
MSL was a bid to strengthen Meta’s position in the race against frontier-AI developers such as OpenAI, Google, and Anthropic. Reuters characterized the move as coming amid competitive pressure and setbacks surrounding Llama 4; that is outside reporting, not a company admission that one release alone caused the reorganization. Reports of senior staff departures and concerns about momentum also formed part of the backdrop.
Zuckerberg personally prioritized recruiting and model development. Meta’s strategic advantage is also its reach: Facebook, Instagram, WhatsApp, Messenger, Threads, Meta AI, and its AI glasses offer many routes to put a model in front of users. Distribution may help adoption, but broad availability is not evidence that the underlying models lead competitors in capability. Reuters’ account of the competitive context described MSL as a response to pressure and Llama 4-related setbacks.
What “personal superintelligence” means
In a July 30, 2025 letter, Zuckerberg framed Meta’s goal as “personal superintelligence”: AI made broadly available to help individuals pursue their goals, with applications in creativity, culture, science, health, and connection. He contrasted that vision with centrally directing superintelligence toward automating all valuable work. That contrast is Zuckerberg’s strategic framing, not an agreed technical distinction. Meta published the letter on July 30, 2025.
- AGI is a contested term commonly used for broadly capable AI at or beyond human level; there is no single universally accepted threshold.
- Superintelligence usually means capabilities exceeding human abilities across essentially all relevant intellectual tasks. Meta’s earnings commentary described it as AI that surpasses human intelligence “in every way.” That is an ambition, not a demonstrated benchmark result.
- Agentic AI can plan, use tools, or carry out multistep tasks. Completing a task with tools does not establish unrestricted autonomy or superintelligence.
- Personal AI is Meta’s product-oriented framing for assistants that use context to help individuals; it is not itself a technical capability standard.
What MSL has produced
Muse Spark and Muse Spark 1.1
On April 8, 2026, Meta introduced Muse Spark as the first model in a new family developed by MSL. Meta described it as natively multimodal, capable of reasoning and tool use, designed for visual understanding, and able to coordinate multiple agents. It initially powered Meta AI through the Meta AI app and meta.ai; API access began in private preview for selected partners. These are Meta’s descriptions of its model. Meta’s Muse Spark announcement and its technical post provide the company’s details.
Meta announced Muse Spark 1.1 on July 9, 2026. Its AI site also identifies Muse Image and Muse Video as MSL’s first media-generation models. The available material does not establish that these models share Muse Spark’s access terms or availability, so they should be treated as separate releases rather than interchangeable versions of one product. Meta’s AI site lists the model family and products.
How Muse relates to Llama and Meta AI
| Name | What it is |
|---|---|
| MSL | Meta’s AI organization. |
| Llama | Meta’s existing model family and ecosystem. Meta’s 2025 remarks said it was progressing on Llama 4.1 and 4.2 while developing next-generation models. |
| Muse | A newer model family developed by MSL, including Muse Spark and media-generation models. |
| Meta AI | The consumer-facing assistant that uses Meta models and is being integrated across Meta products. |
| Meta Model API | A route for developers to access models; availability and terms differ from consumer access. |
Meta has not established that Muse replaces Llama. Nor should “open” be used as a blanket synonym for open source: a release can expose weights, code, or an API without publishing training data or every component. Check the specific model’s release terms before relying on it for development.
What Meta AI can do, and where it is rolling out
Meta said on July 24, 2026 that Meta AI powered by Muse Spark 1.1 was gaining more task-oriented features in selected markets. The company described daily briefings, research and information synthesis, connections to email and calendars, slide decks, recurring tasks, mood boards and plans, work users can steer while it is underway, saved work, and shopping-related discovery. Meta said the assistant could use content and recommendations from its apps. These features were announced as a staged rollout, not as universally available functionality. Meta’s July 24 announcement describes the features and rollout.
Meta’s deployment strategy spans the Meta AI app and meta.ai, with integrations across WhatsApp, Instagram, Facebook, Messenger, Threads, and Ray-Ban Meta and Oakley Meta glasses. Rollout differs by country and product surface; Meta cited the United States and Canada for portions of the glasses rollout and said additional countries and surfaces would follow. A feature being announced, available in one app, or in a limited market does not mean it is active for every user or powered by the same model version.
Best Value
For developers, Meta’s model-access page described Muse Spark as in public preview for US developers and said accounts received $20 in free credits at launch. That one-time launch credit is not a recurring allowance or a full long-term pricing schedule. Meta also listed access through OpenRouter, but the cited information does not establish OpenRouter’s current Muse Spark price. See Meta’s developer and model-access page for the stated access terms.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What Meta is investing—and what reported figures mean
Meta’s bet involves several distinct kinds of investment; they should not be collapsed into a single “cost of MSL.”
- Scale AI: Reporting described a $14.3 billion investment for a 49% stake, alongside Wang’s move to Meta. That is not the same as acquiring Scale AI outright. Axios covered the investment and leadership move.
- Infrastructure: Reuters reported Zuckerberg’s longer-term ambition to invest hundreds of billions of dollars in AI data centers. That is a long-range plan, distinct from annual capital-expenditure guidance or money already spent. Reuters reported on the infrastructure ambition.
- Recruiting: Reports described high-value offers to researchers and engineers from other AI organizations, with some headline figures reaching hundreds of millions of dollars over multiple years. Such figures may refer to potential total compensation, including equity, rather than guaranteed cash, a signing bonus, or an amount already paid. Reuters’ talent-recruitment factbox reports the offers.
For investors, the important distinction is between committed capital spending, long-term infrastructure ambitions, equity investments, and compensation offers. The cited reporting does not provide one verified total cost for MSL.
What remains unproven
Meta has demonstrated product releases and is deploying more capable assistants, but the announcements cited here do not independently establish that Meta has achieved superintelligence or that MSL leads the field. Company-selected benchmark comparisons need to be evaluated benchmark by benchmark, including test version and prompting conditions, and ideally independently reproduced; a headline comparison alone is not decisive.
Task execution also creates practical risks. Research can include fabricated claims or citations; calendar and email actions can be wrong; performance can vary by language, market, app, or model version. Connecting an assistant to personal accounts raises questions about permissions and data handling, while camera-equipped glasses add privacy concerns for both wearers and bystanders. Meta’s product descriptions are not a substitute for checking available controls and testing behavior before granting access or relying on output. Health-related assistance should not be treated as diagnosis or professional care, even though Meta says it consulted physicians on related capabilities.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
