xAI quietly launches Grok 4.1, a more accurate and emotionally intelligent AI model

Introducing Grok 4.1: xAI's New Model Promises Deeper Understanding and Emotional Intelligence

In a move that has significant implications for the future of AI interaction, xAI has quietly rolled out its latest language model, Grok 4.1. This new iteration promises a leap forward in how AI understands and responds to human users, boasting enhanced accuracy and a newly cultivated emotional quotient. Available across grok.com, the X social platform, and dedicated iOS and Android applications, Grok 4.1 seamlessly integrates into user experiences, appearing automatically in Auto-mode or selectable manually.

A Significant Leap in Performance and Nuance

The xAI team asserts that Grok 4.1 is far more attuned to the subtle intentions behind user queries. It excels in handling tasks that require emotional depth, creative flair, and collaborative synergy. The model's responses are reported to be more consistent and natural, mimicking human conversation with greater fidelity. This enhanced capability is underscored by its impressive performance on LMArena Text Arena, a crucial benchmark for evaluating AI language models. Grok 4.1's 'Thinking' variant (powered by quasarflux) achieved an outstanding Elo score of 1483, surpassing all other non-xAI models by a remarkable 31 points. Even the faster 'Tensor' version secured a strong second place with 1465 Elo, outperforming the full reasoning configurations of other prominent models. In a surprising turn, Gemini 2.5 Pro, a leading competitor, landed in third place, highlighting Grok 4.1's significant advancements. This is a stark contrast to its predecessor, Grok 4, which languished in 33rd position.

Delving into Emotional and Creative Aptitude

To quantify its newfound emotional intelligence, Grok 4.1 was rigorously tested on EQ-Bench3, a challenging dataset comprising 45 multi-turn conversational scenarios designed to assess empathy, emotional comprehension, and interpersonal skills. The results were highly encouraging, with Grok 4.1 receiving top marks both in specific categories and overall normalized Elo rankings during comparative evaluations. Further validation came from creative writing v3, a set of 32 writing prompts evaluated over three iterations. Here, Grok 4.1 demonstrated substantial progress in generating imaginative and compelling creative content.

Reducing Hallucinations for Greater Reliability

A key focus for the xAI development team was to improve the accuracy and reliability of the model, particularly its faster iterations. Significant effort was dedicated to enhancing responses to informational queries and minimizing factual errors, often referred to as 'hallucinations' in AI. Through extensive testing with real-world user queries and the open-source FActScore benchmark – a battery of 500 factual biographical questions – xAI observed a notable reduction in instances of fabricated information. This commitment to factual accuracy is paramount for building user trust and ensuring the model's utility in critical applications.

A Stealthy Deployment and User Validation

The rollout of Grok 4.1 was executed with a degree of discretion, taking place between November 1st and November 14th, 2025. During this period, select groups of users engaged with earlier builds on grok.com, X, and the mobile apps. This beta phase allowed the xAI team to conduct blind comparative tests against the previous model using actual user traffic. The feedback was overwhelmingly positive, with users preferring Grok 4.1's responses in an impressive 64.78% of cases, a testament to its tangible improvements.