Meta's internal codename Watermelon refers to its next frontier large language model after Muse Spark, reportedly trained with an order of magnitude more compute and claimed internally in July 2026 to match OpenAI's GPT-5.5 on unnamed benchmarks. As of early October, the model remains unreleased with no official weights, model card, or independent verification, following a targeted October window cited in internal documents and an absence from the September Meta Connect keynote. Trader sentiment centers on whether Meta will debut it soon on text leaderboards like LMArena amid ongoing pre-training adjustments, heavy infrastructure spending, and competition from advancing models at OpenAI, Anthropic, and Google. Key near-term catalysts include potential integration into Muse services or an open-weight release that could shift arena rankings.