xAI releases Grok 4.1
xAI tuned the update for personality and reliability rather than raw reasoning, reporting a two-week blind test in which users preferred it to Grok 4 64.8% of the time.
- Models & capabilities
- Minor
xAI released Grok 4.1, an update it said reused the reinforcement-learning infrastructure built for Grok 4 but redirected it toward style, personality and emotional intelligence rather than raw reasoning ability. The model ships in a fast conversational variant and a “Thinking” variant, and xAI reported both landed among the top scorers on LMArena’s Text Arena leaderboard at release.
xAI said the model was tested against the previous production Grok in a blind, two-week rollout to real users between 1 and 14 November 2025, in which testers preferred Grok 4.1’s responses 64.8% of the time. The company also reported strong scores on EQ-Bench, a third-party benchmark for emotional intelligence in roleplay scenarios, ahead of Gemini 2.5 Pro, GPT-5 and Claude Opus 4 on that measure, and said hallucination rates had fallen relative to Grok 4. As with most model-release benchmarks, the comparisons were run or commissioned by xAI itself rather than independently reproduced.
The release continued xAI’s compressed release cadence, coming roughly four months after Grok 4, and positioned the model as competing less on the exam-style benchmarks that had dominated earlier frontier-model comparisons and more on the conversational qualities increasingly used to market consumer chatbots.