On November 18, xAI released Grok 4.1 to all users across grok.com, X, and its iOS and Android apps. The company stated that the model was deployed immediately in Auto mode, though it can also be explicitly selected from the model picker. xAI explained that they utilized the same large-scale reinforcement learning infrastructure as Grok 4, applying it to optimize style, personality, utility, and alignment. They also developed a new method using frontier agentic reasoning models as reward models. During a two-week silent rollout, blind pairwise evaluations were conducted on live traffic, where Grok 4.1 recorded a 64.78% win rate over the previous generation model. In the LMArena Text Arena, the reasoning mode (quasarflux) ranked 1st overall with 1483 Elo, while the non-reasoning mode (tensor) ranked 2nd with 1465 Elo.
Sources: Grok 4.1 (HN 140pt, 128 comments) (HN Search (backfill))