Grok 4.20
xAI's multi-agent model that debates internally before answering
Grok 4.20 is xAI’s multi-agent model, with a beta launching February 17, 2026 and the dated release following on March 9. Its defining feature is a native four-agent system, internally named Grok, Harper, Benjamin, and Lucas, that reasons in parallel and debates internally before returning a single answer, which xAI says cut hallucinations by 65 percent compared to Grok 4.1. It runs on a roughly 3 trillion parameter mixture-of-experts backbone with a context window up to 2 million tokens, operates in separate reasoning and non-reasoning modes, and keeps the real-time X data integration xAI has used across the Grok line. API pricing is $1.25 per million input tokens, $0.20 for cached input, and $2.50 per million output tokens, well below the $2/$6 xAI charged for Grok 4.5 seven months earlier.