Meta is preparing to launch a consumer AI agent internally known as Hatch. According to documents obtained by The Information, it could arrive within the next few weeks. The price under consideration for the highest subscription tier is $199.99 per month. That is twenty-five times more than the company charged three months ago for access to its chatbot. A new model codenamed Watermelon is also planned for October.
What Hatch is supposed to do
Meta trained the agent to work with services people use every day. Specifically, these include DoorDash, Etsy, Reddit, Yelp, and Microsoft Outlook. Users can order food, buy products, browse reviews, or sort their email. The whole point is that they will not have to leave the app they are currently using to perform any of these tasks. Early prototypes also include a dashboard featuring tools that the agent has assembled for the user. These could include a fitness tracker or a vacation itinerary.
Hatch is a response to OpenClaw, an open-source personal agent. It runs directly on the user's computer and manages email, calendars, and other services through messaging apps. The tool is free to use. However, you have to set it up yourself, which requires working with the command line and configuring models, credentials, and permissions.
Meta is selling a user interface and adding its own infrastructure to it. This combination defines the entire service, which is precisely why pricing has become the main topic. OpenClaw creator Peter Steinberger himself demonstrated how much such an operation costs. He spent $1.3 million on tokens in a single month while operating agents at scale. Someone has to pay for this computing power, and Meta has decided to pass the costs on to users.
However, the relationship between Meta and OpenClaw has its complications. The company banned its employees from using the tool as early as February 2026, citing security concerns. This happened shortly before Steinberger left for OpenAI.
Who will pay for it, critics ask
The final pricing has not yet been determined. The Information reported back in June on a tiered model under consideration, with the highest tier priced at $199.99 per month offering higher usage limits. Internal plans refer to this tier as Hatch Plus. The version will offer five to ten times the daily capacity of the free option, with limits resetting each billing period. From this perspective, this is not the price of an assistant. It is the price of computing power sold on a monthly basis.
Jeremiah Dillon summarized the objection in a single sentence on X. In his view, the market for a consumer version of Hatch priced at $199.99 per month is exactly zero. Dillon works in product marketing at data company Confluent and previously held positions at Stripe, Google, Amazon, Salesforce, and MuleSoft.
However, he added that The Information tends to be right, so there is probably something behind it that no one on the outside can see. If someone told him that operating Meta's agent costs the company an average of $199.99 per user per month in API fees, he would believe it. Such an interpretation completely changes the meaning of the figure. It ceases to be a price tag and becomes an estimate of operating costs.
WhatsApp app for agents
At the same time, Meta is building a feature in WhatsApp that will allow people to connect agents from other companies and communicate with them. Early access could begin as soon as this week, according to The Information. The company is thus taking on two roles at once. It is selling its own agent while also turning WhatsApp into a platform through which competitors' agents can reach users. The second role is more valuable because distribution will outlast any individual product.
Meta has already begun making money in this area. Since August 1, it has charged two dollars per million tokens for a business agent on WhatsApp.
Built on third-party models
The Information first reported on the project in May. At the time, Hatch was running during development on Anthropic's Claude Opus 4.6 and Claude Sonnet 4.6 models. Meta planned to move it to its own Muse Spark model family before launch.
Watermelon is expected to be next in line, with its release planned for October. The publication was unable to determine whether it will be released under the Muse brand or receive its own name. Internal reports say that the model matched GPT-5.5 in internal testing, thanks to roughly ten times the computing power of Muse Spark.
In any case, the release pace is rapid. Muse Spark 1.2 and the Muse Code terminal agent arrived on August 5. Five days later came Muse Glimmer, an open-weight model with thirty billion parameters. Meta Superintelligence Labs is behind this entire product line. The lab has been led since June 2025 by Alexandr Wang, whom Meta brought in through a $14.3 billion acquisition.
Why Meta needs this to succeed
In the second quarter, Meta reported revenue of $60.8 billion. Advertising accounted for $59.36 billion of that amount, representing approximately 98 percent. The company estimates this year's capital expenditures at $130 billion to $145 billion, including finance lease payments, with the midpoint of the range nearly double last year's figure.
Zuckerberg explained the rationale behind this spending during the July 29 earnings call. He described three areas toward which the computing power is being directed. The first is maintaining advertising and recommendation systems. The second is expanding business agents and developer interfaces, and the third is deploying consumer agents for 3.5 billion users. In his view, it would be a mistake to simply sell all the computing power and gain only a short-term profit.
On August 10, he published an essay on the subject. Meta will build personal agents for billions of people, offer free versions, and charge users who require more computing capacity. Hatch Plus, with five to ten times the free limit, is precisely this plan translated into a pricing structure.
Sources: thenextweb.com and finance.yahoo.com
