Meta's Watermelon model, the unreleased frontier large language model from its Superintelligence Labs, has drawn trader attention due to internal claims of benchmark parity with OpenAI's GPT-5.5 alongside a reported order-of-magnitude increase in training compute over the prior Muse Spark release. Alexandr Wang highlighted the progress in a July 2026 town hall while the model remained in pre-training, with an internal October 2026 target cited in August reporting. No weights, API access, or arena evaluation appeared at the September 23 Meta Connect event, though recent updates note accelerated integration plans and security hardening for related Muse services ahead of broader deployment. Traders are watching for any verified public text arena appearance or official release timeline in the coming weeks.