Meta’s reported progress on its internal “Watermelon” frontier large language model centers on July 2026 statements from AI chief Alexandr Wang that the still-training system matches OpenAI’s GPT-5.5 on unnamed internal benchmarks while using roughly 10x the compute of the April-launched Muse Spark series. Credible reporting from The Information and others points to an October 2026 public release target tied to Meta’s $125–145 billion infrastructure spend and plans to power the new Muse agent platform. No launch occurred at the September 23 Meta Connect event, however, and OpenAI has since previewed GPT-5.6. Traders are watching for official announcements, safety reviews, or further delays that could shift the market-implied odds around the October–November window.