Grok 4.1: Enhanced Emotional Intelligence and Creative Capabilities Roll Out Across Platforms

Grok 4.1 is now available to all users on grok.com, 𝕏 (formerly Twitter), and the iOS and Android apps. The model is rolling out immediately in Auto mode and can be explicitly selected as “Grok 4.1” in the model picker.

Silent Rollout: November 1–14, 2025

Before its official release, Grok 4.1 underwent a silent rollout from November 1 to 14, 2025. During this period, preliminary builds were gradually introduced to an increasing share of production traffic across grok.com, 𝕏, and mobile applications. Continuous blind pairwise evaluations were conducted on live traffic to ensure the model's performance met our high standards.

Significant Improvements in Real-World Usability

Grok 4.1 introduces notable enhancements in creative, emotional, and collaborative interactions. Here are the key improvements:

• Enhanced Perception of Nuance: Grok 4.1 is more perceptive to nuanced intent, making it better at understanding subtle cues in conversations.
• Compelling Conversationalist: The model is more engaging and compelling to interact with, enhancing user experience.
• Coherent Personality: It maintains a coherent personality, ensuring consistent and reliable interactions.
• Retained Intelligence and Reliability: All the razor-sharp intelligence and reliability of previous models are fully retained.

Technical Achievements

To achieve these improvements, we leveraged our large-scale reinforcement learning infrastructure from Grok 4. We optimized the model's style, personality, helpfulness, and alignment using non-verifiable reward signals. Here’s a breakdown of the technical approach:

• Frontier Agentic Reasoning Models: These models were used as reward models to autonomously evaluate and iterate on responses at scale.
• Continuous Optimization: The silent rollout allowed for continuous optimization based on real-world interactions.

Performance Metrics

Compared to its predecessor, Grok 4.1 is preferred in blind evaluations:

• 64.78% Preference Rate: In live traffic, users prefer Grok 4.1 over the previous production model 64.78% of the time.

State-of-the-Art General Capability

Grok 4.1 sets a new standard in blind human preference evaluations, as evidenced by its performance on the LMArena Text Leaderboard:

• Top Overall Position: Grok 4.1 Thinking (code name: quasarflux) holds the #1 overall position with 1483 Elo, leading by 31 points over the highest non-xAI model.
• Non-Thinking Mode Excellence: Grok 4.1 in its non-reasoning mode (code name: tensor), which uses no thinking tokens for immediate responses, ranks #2 at 1465 Elo. This version surpasses every other model’s full-reasoning configuration on the public leaderboard.
• Significant Improvement Over Grok 4: Grok 4.1 significantly outperforms Grok 4, which had an overall rank of #33.

Emotional Intelligence

Grok 4.1 excels in emotional intelligence, a critical aspect for human-like interactions:

• Measurable Improvements: The model demonstrates better understanding and response to emotional cues, enhancing its ability to engage in empathetic and supportive conversations.

Conclusion

The release of Grok 4.1 marks a significant step forward in AI models' real-world usability, particularly in creative, emotional, and collaborative interactions. With enhanced perception, compelling conversational skills, and coherent personality, Grok 4.1 sets a new standard for AI performance.