Meta's internal codename Watermelon refers to the next frontier-scale large language model under development at its Superintelligence Labs, positioned as a successor to the Muse Spark series released in April. Reports from July indicate the model, trained on roughly 10x the compute of its predecessor, reached parity with OpenAI's GPT-5.5 on unnamed internal benchmarks per Chief AI Officer Alexandr Wang, though no independent verification exists. Internal documents cited in August targeted an October 2026 release, yet the model did not appear at Meta Connect in September and training pauses have introduced delays. Trader sentiment reflects Meta's aggressive infrastructure spending alongside uncertainty over exact timing, distribution model, and whether it will ship open weights or via API. Key upcoming catalysts include any official announcements or capability demos that could shift the closely watched October window.