xAI has iterated rapidly on its Grok 4 series throughout 2026, with Grok 4.7 launching in September featuring a 500k-token context window, configurable reasoning levels, and improved agentic performance. These updates target frontier reasoning on demanding benchmarks like Humanity’s Last Exam, a 2,500-question expert-level test designed to resist saturation. However, current Grok variants trail leaders such as Claude Fable 5.1 and Opus 5.5, which post 59–65% scores across verified leaderboards, while Grok entries sit in the low-to-mid 40s. Traders will watch for any Grok 5 preview or major capability jump before year-end, alongside broader shifts in training data, scaling laws, and benchmark refinements that could alter relative positioning among closed frontier models.