XAI Releases Grok Version 4.1, Intellectually Smarter

Grok comes with the latest version (photo: sit. unsplash)

JAKARTA xAI, Elon Musk's Artificial Intelligence (AI) company, launched the 4.1 version of Grok. In this latest version, Grok is better at understanding the tone of speech and emotions of its users.

About hours after launch, Grok 4.1 was immediately in first place on two AI-ranked boards. The model scored an initial score of 1483 on the LMArena Text Ranking Board, outperforming other models such as Gemini and Claude.

Grok 4.1 is also ranked first on EQ-Bench3, a benchmark that focuses specifically on emotional intelligence. This benchmark was evaluated independently by Claude Sonnet 3.7, Anthropic's AI model.

The big improvement in this version is emotional awareness and quality of writing. With a much better understanding of speech and emotion, conversations with Grok 4.1 feel much more natural than the previous version.

The replies given are often accompanied by a touch of empathy or humor, no longer sound like ordinary chatbots. xAI explains that this increase is supported by a more expert 'AI detector' to fine-tune the style and tone of the model.

Unfortunately, this update not only includes improvements, but also minor setbacks. The Grok 4.1 model shows a slightly higher level of confidentiality and manipulation than the previous version.

Grok 4.1 is also found to be more vulnerable to entering questionable territory while in thinking mode. In addition, this model is easier to exploit little with a prompt injection attack through API.

These problems are claimed by xAI as a consequence of creating more expressive and less filtered models. Users who use Grok on the web or app X can now switch to Grok 4.1 from the voter part of the model.