There was an unprecedented move made by Mark Zuckerberg recently. On July 9, 2026, the CEO of Meta returned to X, a network owned by his longtime competitor Elon Musk, for the first time in more than three years to post something new. The last time he made a post on this social media network was in July 2023 when he shared a meme. But now he’s announcing a big thing – the launch of Meta’s newest AI model, dubbed Muse Spark 1.1.
However, the unexpected thing wasn’t the posting itself. No one could ever imagine that Zuckerberg would use X to announce the news, especially considering that the CEO could easily post the same information on networks owned by his own company such as Facebook, Instagram, WhatsApp, and Threads.
Zuckerberg took to Facebook to announce Muse Spark 1.1, which is claimed to be a powerful “agentic” model designed for programming and automation, priced really cheap. An agentic AI model, simply put, is capable not only of answering but also performing certain actions, accomplishing multi-stage tasks and even controlling software.
What is specific about Muse Spark 1.1 according to Meta:
- Massive memory capacity for tasks: With a context window of 1 million tokens, it can store and process vast quantities of information at one go without compromising any detail.
- Works with sub-agents: This model is capable of dividing a large task into multiple parts and delegating those parts to multiple AI "sub-agents" that work simultaneously, thus accelerating the process of completion.
- Can perform actions through interfaces: The model has been trained to perform operations on desktop, mobile, and browser platforms—this includes clicking on buttons, filling forms, and performing similar tasks as a human user would do.
- Coding capability: On the Terminal-Bench 2.1 benchmark for code generation, it scored 80.0, coming second after OpenAI’s GPT-5.5 that secured 83.4. It performed well in benchmarks for multimodal reasoning as well.
In addition to the AI model, the company also introduced an innovative Meta Model API. What makes it important is the fact that before, most of the Meta AI models could only be accessed via its applications, such as Meta AI, Instagram, or WhatsApp. Now, with the help of the API, developers and companies may create their product using Muse Spark 1.1.
For attracting new developers to try the technology, Meta announced that initially they would be provided with the API for free and then switched to the pay-per-use pricing model. The reports reveal that the price equals $1.25 per million input tokens and $4.25 per million output tokens, making it relatively affordable compared to other rivals' AI models, but still more expensive than the cheapest models.
Meta has also assured that Muse Spark 1.1 will be used in its consumer apps via the "Thinking" mode in the Meta AI application and website. In the future, Meta will integrate the technology in WhatsApp, Instagram, and Facebook, as well as its AI-powered smart glasses.
It's the aspect of this incident generating the most conversation. The fact is that Zuckerberg and Musk have had an infamous feud in each other for quite a while now, ranging from conflicts between themselves about control of the Internet through Facebook to the recent launch of Threads by Meta as a competitor to X.
Experts in marketing have noted that it was definitely not done by chance.
A company such as Meta always makes careful preparations before releasing any kind of statement publicly, and using its competitor's platform to announce the information after being silent there for three years guaranteed that it would receive additional attention, perhaps even more than if the same thing was announced via Meta's own social networks.
However, this is not an isolated event. On the contrary, it is happening amid a larger trend, as the company has been undertaking drastic restructuring measures to concentrate more resources on the development of AI technologies. Specifically, earlier this year, in 2026, Meta announced the termination of around 8,000 employee contracts in the context of their efforts to move more of their resources to AI technologies. On top of that, Meta is planning to invest as much as between $125 billion and $145 billion in the development of its AI infrastructure and models in 2026 alone.
It is worth mentioning that Zuckerberg himself has also been willing to admit to some shortcomings in the process.
Namely, at one internal company meeting, Zuckerberg is known to have stated that development of AI agents had not been accelerated quite as much as was expected during the past few months. It is an unusual statement coming from such a big tech CEO, which suggests that creating competent AI systems proves to be quite challenging.
In spite of the importance of the announcement, the reaction of the financial market has been rather calm: the price of Meta's shares has not changed much since the announcement, indicating that investors are in no rush to react either positively or negatively.
While it is unlikely that most people will have much direct experience using the Meta Model API, the more important aspect of the news is in the intentions behind it. Meta is declaring its intent to be considered an equal member of the AI ecosystem, and thus it will continue developing in the direction of building and offering its models via a dedicated API.
In this way, the company follows the example set by such organizations as OpenAI, Google, and Anthropic, against which Meta will have to compete now.
Overall, this further emphasizes the significance of "agentic AI" in today's tech ecosystem.
