According to internal documents reviewed by (The Information), Meta plans to launch, within just a few weeks, a consumer version of an AI agent product codenamed OpenClaw; it also sets a target for October to roll out a new AI large model codenamed “Watermelon (Watermelon).”
Hatch carries a key objective for Meta CEO Mark Zuckerberg: to commercialize the company’s massive investment in artificial intelligence so that revenue sources are no longer overly dependent on advertising. (The Information) previously reported that Meta has considered adopting a tiered pricing model, under which the monthly subscription fee for the premium version could be as high as $199.99, with higher-priced tiers offering greater usage limits.
After training, Hatch can access various websites including DoorDash, Etsy, Reddit, Yelp, and Outlook. Based on early prototypes, it comes with a customizable control panel that can display tools or functional modules generated by the agent—such as a fitness tracker or a travel itinerary planning tool. According to the files, Hatch is expected to launch officially in late August to early September.
Meta is also preparing to launch a platform on WhatsApp that will let users connect other third-party AI agents, and hold conversations with various AI systems directly within the messaging app. A limited group of users is expected to get access to the phased rollout as early as this week.
Meta’s spokesperson declined to comment.
In recent months, Meta has clearly accelerated the pace of AI model releases and iterations, hoping to close the gap with competitors such as OpenAI and Anthropic. In April this year, Meta’s Super/Intelligent Lab released its first model, Muse Spark, positioning it as a multimodal reasoning model, with a focus on tool calling and multi-agent collaborative execution.
In July, Meta released an upgraded version of Muse Spark, 1.1, with an emphasis on strengthening code-writing, computer operation, and other agent-task capabilities. In August, it released Muse Spark 1.2 and simultaneously launched the code agent for software engineering tasks, Muse Code. This sequence of releases indicates that Meta is moving more actively into the arena of coding capabilities and AI agents—an area that has already become a primary battleground for OpenAI and Anthropic.
It is still unclear whether the Watermelon model will be classified under the Muse model series, or whether it will be released externally as an entirely new standalone model.
Meanwhile, Meta continues to adhere to an open-source AI route. Earlier this month, Meta released a relatively lightweight open-source model, Muse Glimmer. CEO Mark Zuckerberg also publicly called on the United States to reduce institutional barriers to the development of open-source AI. Meta also announced it will release an open-weights version of Muse Spark 1.2.
Meta is pursuing a dual-track strategy: on the one hand, it is quickly developing its own proprietary closed-source models and AI agents to match the closed-source products of OpenAI and Anthropic; on the other hand, it continues to externally release open-source large models.