xAI launches Grok 4.1 with improved accuracy and emotional understanding
What's the story
xAI has launched Grok 4.1, the latest iteration of its large-language model (LLM). The upgrade goes beyond just speed and minor tweaks, focusing on real-world conversations with improved style, personality, and coherence. The company claims that this version is "exceptionally capable in creative, emotional, and collaborative interactions," while retaining the sharp reasoning and reliability of its predecessors.
Training techniques
Grok 4.1's advanced training and evaluation methods
To develop Grok 4.1, xAI leveraged its large-scale reinforcement-learning infrastructure (used for Grok 4) and introduced new reward-model systems. These systems use frontier agentic reasoning models to autonomously evaluate and refine responses on a massive scale, optimizing style, personality, helpfulness, and alignment. During a silent rollout from November 1-14, 2025, xAI exposed the model to live traffic and ran continuous blind pair-wise evaluations.
Performance metrics
Grok 4.1 outperforms its predecessor in evaluations
Grok 4.1 was preferred 64.78% of the time over the previous production model during these evaluations. On the LMArena Text Leaderboard, its reasoning version (code-named quasarflux) scored an Elo score of 1483, well ahead of non-xAI models. Even the non-reasoning variant (tensor) hit an Elo score of 1465, ranking #2 and surpassing many full-reasoning competitors in the process.
Enhanced capabilities
Grok 4.1's emotional intelligence and creative writing skills
Grok 4.1 also excels in emotional intelligence (EQ-Bench3) and creative writing (Creative Writing v3). For instance, when responding to emotionally heavy prompts like "I miss my cat so much it hurts," the new model provided a noticeably deeper, more empathetic response than its predecessor. This shows how much xAI has focused on making Grok 4.1 not just a faster AI but one that understands human emotions better too.
User experience
Grok 4.1's practical improvements and availability
xAI emphasizes that Grok 4.1 is not just a lab-tested model but one that delivers improved emotional awareness, tone, and coherent personality across interactions in real-world scenarios. It also reduces hallucinations by 3x (especially in information-seeking contexts) by improving methods for detecting and minimizing factual error. The new model is now live for all users on grok.com as well as iOS and Android apps, starting with Auto mode with an option to explicitly select "Grok 4.1" in the model selector.