Meta's Hatch Goes Live in Weeks: Watermelon Model Slated for October as the Agent Platform Takes Shape

After three months of leaks, Meta’s consumer agent has a launch window. According to internal documents seen by The Information and reported on August 24-25, 2026, Meta plans to release Hatch, its OpenClaw-inspired consumer AI agent platform, in the coming weeks — with late August or early September named as the target. Alongside it comes a denser pipeline than anything Meta has shipped before: a WhatsApp integration that lets users plug third-party AI agents into the messaging app starts trialing within days, and a new frontier model codenamed Watermelon is slated for October.

Taken together, the documents sketch the most concrete picture yet of how Mark Zuckerberg intends to convert the company’s enormous AI capital expenditure — $125 billion to $145 billion committed for 2026 — into revenue that does not depend on advertising. A Meta spokesperson declined to comment on the internal documents, so the details below rest on The Information’s reporting rather than official confirmation.

Hatch: from viral side project to platform bet

Hatch is the consumer-facing descendant of OpenClaw, the agentic framework that became a grassroots sensation after developer Peter Steinberger released it in November 2025. The lineage is famously ironic inside Meta: the company banned internal use of OpenClaw in February 2026 over security concerns, shortly before Steinberger departed for OpenAI — and then built its consumer agent strategy on the same playbook.

What the new documents add is specificity about what Hatch actually does at launch. The agent is trained to navigate and interact directly with external services — DoorDash, Etsy, Reddit, Yelp, and Microsoft Outlook are all named — completing tasks end to end rather than merely chatting about them. Prototype designs feature a customizable dashboard where agents surface “skills” they have built for the user, with examples including automated fitness tracking and travel itinerary planning. The design reads less like a chatbot and more like an operating system for delegated errands.

Monetization follows the tiered structure flagged in earlier reporting: subscriptions reaching $199.99 per month for a premium tier with expanded usage capacity, with lower tiers expected closer to the $20-and-under range that ChatGPT Plus and Claude Pro occupy. If those numbers hold at launch, Hatch would become Meta’s first paid consumer AI product — and one of the most expensive consumer AI subscriptions anywhere.

Watermelon: October’s frontier bet

The second headline in the documents is the model roadmap. Meta is targeting an October release for its next frontier model, codenamed “Watermelon” — though the internal documents reportedly do not clarify whether it will be folded into the existing Muse family or ship as a standalone system.

The codename has been circulating since July, when Meta Superintelligence Labs chief data officer Alexandr Wang told colleagues the in-training model had caught up to GPT-5.5 on internal benchmarks, and that it was trained with roughly an order of magnitude more compute than its predecessor. Watermelon is the direct successor to Muse Spark — the April frontier release Meta internally calls “Avocado” — and its July update, Muse Spark 1.1, which Meta positioned as a multimodal reasoning model built for agentic tasks.

Those performance claims remain internal and unverified, and Meta’s recent track record counsels patience: the much-anticipated Llama 4 Behemoth has still not shipped more than a year after announcement. But the cadence has visibly quickened — Muse Spark in April, Muse Spark 1.1 in July, Muse Spark 1.2 and the Muse Code agent in August, plus the open-weights 30B Muse Glimmer — and Watermelon in October would extend the streak into the fall.

The WhatsApp agent platform: an open lane for third parties

Perhaps the most strategically under-discussed item in the report is the WhatsApp integration. Meta is preparing to let users connect with third-party AI agents directly inside WhatsApp, with a limited user trial beginning as early as this week. Yahoo Finance’s coverage notes the feature has already entered limited trials as a precursor to Hatch itself.

This is a different kind of move than a first-party assistant. A WhatsApp agent platform turns the world’s largest messaging app into a distribution channel for other people’s agents — closer to an app store moment than a chatbot launch. For businesses already paying for the WhatsApp Business Agent (which began charging $2.00 per million tokens on August 1, roughly $0.04-0.05 per typical exchange), it extends the monetization surface from business messaging into agentic services. And it positions Meta to take a platform cut on an ecosystem, not just sell one agent of its own.

Why the timing is forced

Zuckerberg laid out the strategic frame on the Q2 2026 earnings call on July 29, describing a three-bucket strategy: maintain core ads and recommendation systems, scale business agents and APIs, and deploy consumer agents to the company’s 3.5 billion users. Defending the capex commitment, he argued it would be “foolish to basically just sell all of the compute and take a short-term profit.”

The pressure to show returns is real. The AI buildout helped drive a steep profit decline in Meta’s most recent quarter, and META stock is down roughly 15% year to date while the company races OpenAI and Anthropic — both of which have their own agent ambitions and, in Anthropic’s case, a potential $2 trillion IPO fueling the comparison. Hatch shipping before Q4 gives Meta a monetization story to tell before year-end results land.

Execution risks remain substantial. The $199.99 price point is a high bar for a consumer product whose category has a short history of horror stories — agents acting after access revocation, credential handling mistakes, prompt-injection attacks through the very web pages they browse. Hatch’s reported training in simulated replicas of DoorDash, Etsy, and Outlook mitigates the cold-start problem but cannot fully close the sim-to-real gap on an adversarial internet. And relying on a single report means details — pricing especially — may shift before launch.

What to watch

Three dates now matter more than any benchmark score: whether the WhatsApp third-party agent trial actually opens this week; whether Hatch appears by mid-September at the reported tiers; and whether Watermelon lands in October with independent evaluations that support the GPT-5.5-parity claims. If all three arrive on schedule, Meta will have moved from chasing the frontier to packaging it as a product — a transformation the company has spent roughly $130 billion this year trying to buy.