Meta is preparing to launch Hatch, a consumer-facing AI agent, as soon as late August or early September, according to internal documents reported by The Information. The most expensive subscription tier would cost up to $199.99 a month in exchange for higher usage limits and advanced capabilities. Meta has not publicly confirmed the product, its launch date, or its pricing.
Hatch is not a chatbot with a new coat of paint. It is designed to act — navigating third-party services on the user’s behalf, chaining steps across apps, and finishing multi-step jobs. Reported test environments include DoorDash, Etsy and Outlook. The distribution is the part nobody else can match: Hatch is being built into Instagram and WhatsApp.
Then there is the detail buried in the reporting. During development, Hatch ran on Claude Opus 4.6 and Claude Sonnet 4.6 — Anthropic’s models. Meta’s own next-generation model for the job, codenamed Watermelon, is targeted for October.
The price is the strategy
Meta has never sold consumers a subscription that mattered to its P&L. The company monetises billions of people in aggregate, a few dollars at a time, through advertising. A $199.99 monthly tier is a completely different business: a small number of paying customers with high expectations and a support burden, rather than an enormous number of free ones.
That price sets the bar for the product. An agent charging $200 a month has to reliably save its buyer more than $200 a month — in time, in errors avoided, or in work not outsourced elsewhere. Agents are not yet reliable at that standard on the open web, which is precisely why the reported test surfaces are a short, controlled list rather than “the internet.”
Renting the engine you plan to replace
Building a premium subscription on someone else’s frontier model means your gross margin sits inside a competitor’s price list. Every task the agent completes is inference Meta pays for and Anthropic prices. That is survivable during development and expensive at consumer scale.
Watermelon is the fix, and October is the date to hold Meta to. Ship Hatch in September on Anthropic and swap the engine in October, and the launch is a bridge. Slip Watermelon, and Meta is running a flagship paid product on a rival’s economics through the holiday quarter.
Our take: The headline will be the $200 price. The real disclosure is that the most distribution-rich agent launch in history is arriving on a competitor’s models, weeks before its own are ready. That is a company choosing to be early over being self-sufficient — a reasonable bet when the scarce asset is habit, not capability. Consumer agents will be won by whoever gets into the app people already have open, and Meta has two of those. But it inverts the usual read on vertical integration: for now, Anthropic gets paid whether or not Hatch works, and Meta carries the churn risk on a product it does not fully control. Watch October, not September.
What to watch
- Whether the price survives launch. $199.99 is reported as the top of a tiered structure, not the entry point. The cheapest tier and its usage cap will tell you far more about the unit economics than the headline number does.
- The Watermelon date. October is the stated target for the in-house model. A slip means paying a competitor’s per-token rate on a subscription product through Q4.
- Which surfaces open up. DoorDash, Etsy and Outlook are a curated list. The moment the agent is allowed onto sites that have not agreed to be automated, the failure modes and the legal questions both change.
- What it does with the social graph. An agent grounded in Instagram and Facebook data knows more about the user than any standalone assistant. That is the differentiator and the regulatory exposure, in the same sentence.