Meta's next frontier large language model, internally codenamed Watermelon and positioned as the successor to the Muse Spark family, remains unreleased as of early October 2026 despite an internal target window centered on that month. Reports from August indicate Meta planned an October debut following a July pre-training pause, with the model trained on roughly 10 times the compute of its predecessor and internal benchmarks claimed to reach parity with OpenAI's GPT-5.5 class. Alexandr Wang, head of Meta Superintelligence Labs, highlighted the progress in a July town hall, framing it as competitive catch-up ahead of broader agentic and coding capabilities. No official announcement or weights have appeared at events like Meta Connect in September, though recent coverage notes accelerated integration efforts post-Muse stabilization. Traders are watching for any late-October confirmation or slippage signals that could shift implied probabilities around near-term deadlines.