Grok 4.1 is now available on grok.com and mobile apps, bringing notable enhancements in its usability. The model is designed for better creative, emotional, and collaborative interactions. Improvements stem from applying advanced reinforcement learning techniques to refine its style and personality while maintaining intelligence and reliability. Users can easily select Grok 4.1 in the model picker, and it’s in Auto mode by default. In direct comparisons, Grok 4.1 has been preferred 64.78% of the time over its predecessor in traffic evaluations. It leads the LMArena Text Leaderboard, achieving an Elo score of 1483, which is 31 points ahead of the nearest non-xAI model. It also ranks second in non-reasoning mode with an Elo of 1465, outperforming all other models’ reasoning configurations. Emotional intelligence was measured using EQ-Bench3, where Grok 4.1 demonstrated significant advancements in understanding and empathy, as seen in its responses to emotionally charged prompts. Creative writing capabilities have also been assessed, with Grok 4.1 producing more engaging and nuanced content compared to earlier versions. For instance, its response to a prompt about consciousness reflects a deeper understanding of the human experience. The model also aims to reduce factual inaccuracies, known as hallucinations, particularly in information-seeking contexts. After post-training, Grok 4.1 showed a marked decrease in hallucination rates, indicating a focus on delivering accurate information.