xAI has released Grok 4.1, a new frontier model available to all users on grok.com, x.com, and mobile apps. The update emphasizes conversational intelligence, emotional understanding, and real-world helpfulness while significantly reducing error rates.
- Grok 4.1 (Thinking) ranks first on the LMArena leaderboard with a score of 1483, surpassing the previous Grok 4.1 variant which holds second place.
- Hallucination rates drop from 12% to approximately 4%, and the model scores over 40 points higher than Grok 4Fast released two months ago.
- During a silent rollout from November 1–14, 2025, it was preferred in 64.78% of blind pairwise evaluations on live traffic.
- The release includes three variants: Grok 4.1 Thinking (quasarflux) for step-by-step reasoning, Grok 4.1 Non-Reasoning (tensor) for speed, and Grok 4.1 Fast with integrated search tools.
The model is positioned as the top released option for creative writing and immediate general utility, with xAI noting that future models like Grok 5 will bring further parameter increases in Q1 2026.