Meta’s recently launched Muse Wonder series, internally codenamed Watermelon, drives current trader focus after its October 12–13, 2026 release. The closed-weight models, trained on roughly 10× the compute of prior Muse Spark releases, reportedly reached frontier parity with OpenAI’s GPT-5.5 class on internal benchmarks and now rank just behind Anthropic’s top Claude Opus 5.5 variants on independent leaderboards such as Artificial Analysis. Humanity’s Last Exam (HLE), a demanding 2,500-question expert benchmark where even leading models score below 65 percent, has not yet featured verified Watermelon results. Traders are watching for imminent HLE evaluations or API releases that could establish the model’s debut on this hard test, alongside Meta’s planned agent and coding updates.