xAI’s Grok 4.7, described by Elon Musk as a roughly 2.1-trillion-parameter model positioned near Claude Opus 5.0 level, remains unreleased as of mid-September 2026 despite multiple missed targets. Grok 4.6 shipped on August 12, after which Musk repeatedly projected 4.7 within weeks, then narrowed it to September 12 before citing reinforcement-learning issues with response-length penalties and task persistence. Recent posts have shifted focus to Grok 4.8 (2.5T parameters on a new in-house C++ training stack) entering RL soon, while competitive launches from OpenAI and Meta add pressure. Traders watch for an official xAI model card, API identifier, or benchmarks, as all timing signals so far have come solely from Musk’s X posts rather than verified releases.