STRONG SIGNAL: when the company's own AI chief names the next model before launch

Editorial label: STRONG SIGNAL. In a Bloomberg interview published on September 2, 2026, Meta's Chief AI Officer Alexandr Wang confirmed the company is already working on a larger, more powerful frontier model internally named Watermelon. The line he dropped, almost in passing, is rare: 'we believe that Watermelon will be extremely competitive.' Executives at major labs usually avoid confirming codenames for unreleased products, and the fact that Wang did so publicly is what separates this from a plain anonymous leak.

The backdrop: Meta just proved it can actually compete

The confirmation doesn't arrive in a vacuum. It comes two days after the launch of Muse Spark 1.3, the fourth Muse Spark model in five months, which Meta described as its biggest performance jump so far. Wang said Muse Spark 1.3 is 'competitive' with Anthropic's Claude Fable 5.1 and 'better than' OpenAI's GPT-5.6 Sol at code generation. An independent evaluation from Artificial Analysis scored the model at 62 on its Intelligence Index, trailing only Fable 5.1 and Claude Opus 5, and ahead of every OpenAI model tested. After months of being treated as the frontier race's straggler, Meta finally has a third party result backing up the talk, which lends extra weight to the Watermelon confirmation: this isn't the company's first empty promise of the cycle.

What's confirmed and what's still anonymous sourced reporting

It matters to separate the two layers here. The Watermelon name and the project's existence come straight from a Meta executive's mouth, in an interview with direct attribution. The concrete details, an October target, roughly ten times the compute of Muse Spark, and a launch package alongside a consumer agent platform called Hatch, come from internal documents reviewed by The Information and echoed by Stocktwits, PYMNTS, Forkast News and Techstrong.ai. Meta has not publicly confirmed either the Hatch name or the October timeline, and declined to comment officially when approached by those outlets.

Hatch: the product that would make Watermelon worth it for the average user

The most interesting angle in the unconfirmed reporting is what Meta plans to do with Watermelon once trained. According to the documents, the company is preparing Hatch, a consumer agent built on a lineage tracing back to OpenClaw, the agentic framework that went viral in November 2025. Hatch would reportedly navigate real websites and complete tasks like DoorDash orders, Etsy purchases, Reddit and Yelp interactions, and Outlook email management, all on the user's behalf. That would reposition Meta from chatbot vendor to digital agent provider, monetizing part of its 3.5 billion user base with a premium subscription the documents peg at up to $199.99 a month, a figure also not officially confirmed.

The track record working against a rigid timeline

Meta has missed AI deadlines before. Behemoth, this generation's spiritual predecessor, was reportedly delayed after the company's own leadership questioned its capabilities at the time. That doesn't invalidate the October target for Watermelon, but it's the reason the specific date should be treated as an internal document's projection, not a public Meta commitment.

MaxAssistant's read

What changes here isn't the catchy name, it's the pattern: for the first time this cycle, a senior Meta executive has publicly confirmed, on the record, that a model project beyond the Muse Spark line exists, is named, and is in active development. Combined with an independent benchmark result that finally puts Meta among the market's top three, the signal is that the company is positioning to fight at the top floor of the frontier, not just in low cost models. The caution belongs on the speed of the story: until Meta publishes a date, a model card, or a pricing page for Watermelon, the rest of the package, Hatch included, remains anonymous sourced reporting about a company that has missed similar promises before.

Sources

SiliconANGLE, Meta says it has caught up with Anthropic and OpenAI with Muse Spark 1.3: https://siliconangle.com/2026/09/02/meta-says-it-has-caught-up-with-anthropic-and-openai-after-releasing-muse-spark-1-3-its-most-powerful-llm-so-far | Tech Insider, Meta Hatch AI Agents, Watermelon Model: https://tech-insider.org/meta-hatch-ai-agent-watermelon-model-2026