The open-source AI movement has never lacked for options. Mistral, Falcon, and a growing field of open-weight models have been available to developers for years. But when Meta threw its weight behind Llama, something shifted. A company with three billion users, vast compute resources, and the credibility of a tech giant was now building openly, and the developer community responded.
By early 2026, the Llama ecosystem had reached 1.2 billion downloads, averaging about 1 million per day. That is the context for what happened on April 8, 2026. Meta launched Muse Spark, its first major new Meta AI model in a year, and the first product from its newly formed Meta Superintelligence Labs.
It is capable in ways Llama 4 never was, benchmarks well against the current frontier, and is completely proprietary. No free download. No open weights. No building on it unless Meta decides you can.
The companyspentUS$14.3 billion, brought in Alexandr Wang from Scale AI to lead its AI rebuild, then spent nine months tearing down its entire AI stack and starting over. Muse Spark is what came out the other side. The developer community that made Llama what it was is now being asked to wait for a future open-source version that may or may not arrive on any predictable timeline.
Muse Spark is a natively multimodal reasoning model with tool-use, visual chain of thought, and multi-agent orchestration built in. It now powers Meta AI, which reaches over three billion users in Meta’s apps. Meta rebuilt its technology infrastructure from scratch, letting the company create a model that is as capable as its older midsize Llama 4 variant for an order of magnitude less compute.
That efficiency number is worth noting. At the scale Meta operates, compute costs compound fast, and running a frontier-class Meta AI model at a fraction of the cost of its predecessors changes the economics of deploying it in billions of interactions daily.
On benchmarks, the picture is genuinely mixed. Muse Spark scores 52 on the Artificial Intelligence Index v4.0, placing it fourth overall behind Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Meta has not claimed to have built the best model in the world, which is itself a departure from the over-claiming that damaged Llama 4’s credibility.
Where Muse Spark leads is health. On HealthBench Hard – open-ended health queries – it scores 42.8, substantially ahead of Gemini 3.1 Pro at 20.6, GPT-5.4 at 40.1, and Grok 4.2 at 20.3. Health is a stated priority for Meta; the company says it worked with over 1,000 physicians to curate training data for the model.
Muse Spark also offers three modes of interaction: Instant mode for quick answers, Thinking mode for multi-step reasoning tasks, and Contemplating mode, which orchestrates multiple agents’ reasoning in parallel to compete with the most demanding reasoning modes from Gemini Deep Think and GPT Pro.
This is the part of the Muse Spark story that the benchmark tables do not capture. Unlike Meta’s previous models, which were released as open-weight models – meaning anyone could download and run them on their own equipment – Muse Spark is entirely proprietary. The company said it will offer the model in a private preview to select partners through an API, making Muse Spark even more proprietary than the paid models offered by Meta’s rivals.
Wang addressed the change directly, stating: “Nine months ago, we rebuilt our AI stack from scratch. New infrastructure, new architecture, new data pipelines. This is step one. Bigger models are already in development with plans to open-source future versions.”
The developer community’s response has been sceptical. Some see this as a necessary pivot after Llama 4 failed to gain expected traction. Others view it as Meta closing the gates once it has something worth protecting. That is the community now being asked to wait while competitors without that open-source legacy continue shipping freely available weights.
Meanwhile, Meta is not waiting for the developer community to come around. Muse Spark will debut in the coming weeks inside Facebook, Instagram, WhatsApp, and Messenger, as well as in Meta’s Ray-Ban AI glasses. That rollout path is arguably more consequential than any benchmark result. OpenAI and Anthropic sell to developers and enterprises. Meta deploys directly to over three billion people already inside its apps daily.
Meta’s push into health does raise privacy questions worth watching. Muse Spark users will need to log in with an existing Meta account to use it, and while Meta does not explicitly say personal account information will be used by the AI, the company has generally trained on public user data and has positioned Muse Spark as a personal superintelligence product.
Meta stock rose more than 9% on the day of the launch, a signal that investors read the Muse Spark release as proof that the US$14.3 billion bet on Wang and the nine-month rebuild produced something real. Whether the promised open-source versions actually materialise is a question the developer community will press every quarter. The answer will define how this chapter of Meta’s AI story is remembered.
See Also: The Meta-Manus review: What enterprise AI buyers need to know about cross-border compliance risk
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
The post Meta has a competitive AI model but loses its open-source identity appeared first on AI News.
Why is old exon science getting new traction? What’s unsettling biotech VCs? And who will be the next CEO of PhRMA?
We discuss all that and more on the latest episode of “The Readout LOUD,” STAT’s weekly biotech podcast.
The MCP Dev Summit North America 2026, held on April 2-3 at the New York Marriott Marquis, gathered about 1,200 attendees. Hosted by the Linux Foundation's Agentic AI Foundation, discussions focused on the Model Context Protocol's evolution and enterprise adoption, particularly by Amazon and Uber, emphasizing security, interoperability, and scaling for production.
By Andrew Hoblitzell
Sepehr Khosravi discusses the current state of AI-assisted coding, moving beyond basic autocompletion to sophisticated agentic workflows. He explains the technical nuances of Cursor’s "Composer" and Claude Code’s research capabilities, providing tips for managing context windows and MCP integrations. He shares lessons from industry leaders on shrinking process time beyond just writing code.
By Sepehr Khosravi
Google has released the open-source Colab MCP Server, enabling AI agents to directly interact with Google Colab through the Model Context Protocol (MCP). The project is designed to bridge local agent workflows with cloud-based execution, allowing developers to offload compute-intensive or potentially unsafe tasks from their own machines.
By Robert Krzaczyński
We evolved for a linear world. If you walk for an hour, you cover a certain distance. Walk for two hours and you cover double that distance. This intuition served us well on the savannah. But it catastrophically fails when confronting AI and the core exponential trends at its heart.
From the time I began work on AI in 2010 to now, the amount of training data that goes into frontier AI models has grown by a staggering 1 trillion times—from roughly 10¹⁴ flops (floating-point operations‚ the core unit of computation) for early systems to over 10²⁶ flops for today’s largest models. This is an explosion. Everything else in AI follows from this fact.
The skeptics keep predicting walls. And they keep being wrong in the face of this epic generational compute ramp. Often, they point out that Moore’s Law is slowing. They also mention a lack of data, or they cite limitations on energy.
But when you look at the combined forces driving this revolution, the exponential trend seems quite predictable. To understand why, it’s worth looking at the complex and fast-moving reality beneath the headlines.
Think of AI training as a room full of people working calculators. For years, adding computational power meant adding more people with calculators to that room. Much of the time those workers sat idle, drumming their fingers on desks, waiting for the numbers to come through for their next calculation. Every pause was wasted potential. Today’s revolution goes beyond more and better calculators (although it delivers those); it is actually about ensuring that all those calculators never stop, and that they work together as one.
Three advances are now converging to enable this. First, the basic calculators got faster. Nvidia’s chips have delivered an over sevenfold increase in raw performance in just six years, from 312 teraflops in 2020 to 2,250 teraflops today. Our own Maia 200 chip, launched this January, delivers 30% better performance per dollar than any other hardware in our fleet. Second, the numbers arrive faster thanks to a technology called HBM, or high bandwidth memory, which stacks chips vertically like tiny skyscrapers; the latest generation, HBM3, triples the bandwidth of its predecessor, feeding data to processors fast enough to keep them busy all the time. Third, the room of people with calculators became an office and then a whole campus or city. Technologies like NVLink and InfiniBand connect hundreds of thousands of GPUs into warehouse-size supercomputers that function as single cognitive entities. A few years ago this was impossible.
These gains all come together to deliver dramatically more compute. Where training a language model took 167 minutes on eight GPUs in 2020, it now takes under four minutes on equivalent modern hardware. To put this in perspective: Moore’s Law would predict only about a 5x improvement over this period. We saw 50x. We’ve gone from two GPUs training AlexNet, the image recognition model that kicked off the modern boom in deep learning in 2012, to over 100,000 GPUs in today’s largest clusters, each one individually far more powerful than its predecessors.
Then there’s the revolution in software. Research from Epoch AI suggests that the compute required to reach a fixed performance level halves approximately every eight months, much faster than the traditional 18-to-24-month doubling of Moore’s Law. The costs of serving some recent models have collapsed by a factor of up to 900 on an annualized basis. AI is becoming radically cheaper to deploy.
The numbers for the near future are just as staggering. Consider that leading labs are growing capacity at nearly 4x annually. Since 2020, the compute used to train frontier models has grown 5x every year. Global AI-relevant compute is forecast to hit 100 million H100-equivalents by 2027, a tenfold increase in three years. Put all this together and we’re looking at something like another 1,000x in effective compute by the end of 2028. It’s plausible that by 2030 we’ll bring an additional 200 gigawatts of compute online every year—akin to the peak energy use of the UK, France, Germany, and Italy put together.
What does all this get us? I believe it will drive the transition from chatbots to nearly human-level agents—semiautonomous systems capable of writing code for days, carrying out weeks- and months-long projects, making calls, negotiating contracts, managing logistics. Forget basic assistants that answer questions. Think teams of AI workers that deliberate, collaborate, and execute. Right now we’re only in the foothills of this transition, and the implications stretch far beyond tech. Every industry built on cognitive work will be transformed.
The obvious constraint here is energy. A single refrigerator-size AI rack consumes 120 kilowatts, equivalent to 100 homes. But this hunger collides with another exponential: Solar costs have fallen by a factor of nearly 100 over 50 years; battery prices have dropped 97% over three decades. There is a pathway to clean scaling coming into view.
The capital is deployed. The engineering is delivering. The $100 billion clusters, the 10-gigawatt power draws, the warehouse-scale supercomputers … these are no longer science fiction. Ground is being broken for these projects now across the US and the world. As a result, we are heading toward true cognitive abundance. At Microsoft AI, this is the world our superintelligence lab is planning for and building.
Skeptics accustomed to a linear world will continue predicting diminishing returns. They will continue being surprised. The compute explosion is the technological story of our time, full stop. And it is still only just beginning.
Mustafa Suleyman is CEO of Microsoft AI.
Login