xAI launched Grok 4.1 with higher emotional intelligence and reduced hallucinations

xAI released Grok 4.1, a new update that enhances the AI for real-world conversations. It is now live for all users on grok.com, the X website, and the iOS and Android apps. You can use it right away in Auto mode or pick “Grok 4.1” from the model selector.

The primary goal of Grok 4.1 is to improve how the AI handles creative tasks, emotional exchanges, and teamwork-style interactions. It picks up on subtle hints in what you say, stays consistent in its personality, and remains sharp and reliable. The team kept the high intelligence from earlier versions but made the AI more engaging and human-like in chats.

To build this, xAI used the same large-scale reinforcement learning system that created Grok 4. And trained the model on style, personality, helpfulness, and alignment. A key advance was using advanced reasoning models to evaluate responses on their own. This allowed training on things that are hard to measure, such as emotional tone or creativity, at a massive scale.

From November 1 to 14, 2025, xAI tested early versions of Grok 4.1 quietly on real user traffic. The AI firm increased the share of users seeing it over time. In blind tests where users compared responses without knowing the model, people preferred Grok 4.1 64.78% of the time over the old production version.

On public leaderboards, Grok 4.1 performs at the top. In the LMArena Text Leaderboard, the thinking mode (called quasarflux) scores 1483 Elo and holds first place. It leads the next non-xAI model by 31 points. The non-reasoning mode (called tensor) scores 1465 Elo, takes second overall, and beats every other model’s full-reasoning setup. This is a significant leap from Grok 4.

LMarena benchmark of xAI Grok 4.1

The update also reduces hallucinations—wrong facts in answers. Tests on real user questions show fewer errors. Grok 4.1 performs better on emotional intelligence benchmarks with roleplay scenarios and creative writing tasks that involve multiple rounds.

There are two main modes now. Thinking mode adds reasoning steps for complex problems. Non-reasoning mode gives fast answers without extra tokens.

xAI says this release makes Grok much more useful on a day-to-day basis. It retains the core smarts but adds a better understanding of intent and more natural flow in talks. Anyone can try it now without extra subscriptions.

(source)