Meta's Watermelon model, the internal codename for Meta Superintelligence Labs' next frontier large language model after Muse Spark, remains in training as of late September 2026. Chief AI officer Alexandr Wang's July internal update highlighted parity with OpenAI's GPT-5.5 on unnamed benchmarks, achieved through roughly 10x the compute of its predecessor amid Meta's $125–145 billion infrastructure spend. Reports from August point to an October release target, though the model skipped Meta Connect in September despite a September 2 watermelon emoji tease from Mark Zuckerberg. Traders are watching for verified public benchmarks, open weights, or API access that could confirm competitive positioning against GPT-5.5 and Claude models, with any slippage in the October window likely to shift sentiment amid rapid frontier iteration.