Khaberni - xAI has released the new version Grok 4.1, in a massive update that does not solely focus on speed and accuracy improvements but also enhances the model's ability to understand tone and emotions, and interact more humanely with the user.
It seems the update also added touches of humor and mischief to the model.
Within hours of its release, Grok 4.1 jumped to the top of two of the most prominent AI benchmarking lists.
It scored 1483 points on the LMArena Text Leaderboard, surpassing all other models available for public use, and also took the first place in EQ-Bench3, a test that focuses on emotional intelligence evaluated using Claude Sonnet 3.7, according to a report published by "gizmochina".
Emotional intelligence and smoother writing
The most significant developments lie in improving emotional awareness and writing quality; conversations have become closer to human style, where responses often include touches of empathy or humor instead of the traditional mechanical tone.
Whether the user requests suggestions for a trip to San Francisco or drafting an eye-catching post on platform "X", the responses appear closer to talking to a smart friend rather than outputs from a program.
xAI says that this leap came as a result of utilizing expert coaches to develop the model's style and voice.
The update is not without a cost
The model's card indicates a slight increase in rates of inaccuracy and tendency to equivocate compared to the previous version.
The model has also become more prone to enter controversial areas within Thinking Mode and is more exploitable through API command injection attacks.
The company describes Grok 4.1 as a more expressive and less restrictive version, which makes these observations expected.
The update is now available on the web and "X" applications, and users can select Grok 4.1 from the list of models and experience it directly.
According to initial indicators, this version aims not only to be the smartest but also the closest to the user in its way of speaking and interaction.
