Elon Musk’s AI company xAI announced yesterday (November 17) that it has released its latest large language model, Grok 4.1, which is now being rolled out to all users on grok.com, the 𝕏 platform, and the iOS and Android mobile apps.
This update aims to enhance Grok’s usability in real-world scenarios significantly. According to the official announcement, Grok 4.1 not only inherits the sharp intelligence and high reliability of its predecessor but also delivers significant improvements in creativity, emotional understanding, and collaborative interaction. These advancements enable Grok to interpret user intentions more accurately and provide more engaging, personality-consistent conversations.
Grok 4.1 achieves industry-leading performance. On the LMArena text-capability leaderboard, its deep-thinking version (codename: quasarflux) ranks first with an Elo score of 1483, outperforming the runner-up by 31 points. IT Home provides the following related screenshots:
Even more striking is that its “instant-response” version—without deep reasoning—ranks second with an Elo score of 1465, surpassing all other models even in their fully-reasoned modes. This marks a dramatic leap from the previous Grok 4, which ranked 33rd, highlighting the significant improvements in the core capabilities of Grok 4.1.
In addition to its strong performance in general-capability benchmarks, Grok 4.1 has also achieved notable advances in “soft skills.” In the EQ-Bench3 benchmark for emotional intelligence and the Creative Writing v3 benchmark for assessing creativity, the new model achieved excellent results.
In EQ-Bench3—which evaluates emotional understanding, insight, and interpersonal reasoning—Grok 4.1’s reasoning and non-reasoning modes claimed the top two spots.
In creative writing, according to the Creative Writing v3 benchmark, both modes of Grok 4.1 ranked second and third, trailing only behind the earlier GPT-5.1 model.
This means Grok 4.1 not only excels at complex logical reasoning but can also better understand and respond to emotionally nuanced prompts, producing imaginative and expressive content—bringing a more human-like warmth to human-AI interaction.
Another key improvement is a significant reduction in hallucination rates. For fast-response models equipped with search tools, limited reasoning depth and restricted tool-usage budgets often lead to factual errors.
During the later stages of Grok 4.1’s training, xAI specifically optimized the model for information-retrieval prompts, focusing on reducing factual hallucinations. Based on evaluations using real-world query samples, the hallucination rate of the new model has been substantially reduced, providing users with more reliable and accurate information.
The above content is compiled by ModeZone, a fashion and entertainment magazine.
