xAI's Grok 4.20 Beta 2 Brings 4-Agent Collaboration and Weekly Rapid Learning

xAI has released Grok 4.20 Beta 2, the second major update to its Grok 4.20 line, introducing a native 4-agent collaboration architecture where four specialized AI personas — Grok (coordinator), Harper (research), Benjamin (logic and code), and Lucas (creative) — work simultaneously on complex tasks and synthesize their outputs into a single unified response. The update also delivers measurable improvements to instruction following and a 25% reduction in capability hallucination, alongside more reliable LaTeX rendering and multi-image display. Perhaps most distinctively, Grok 4.20 introduces a rapid learning loop that improves the model on a weekly cadence using real-world feedback — a first for any frontier model.