X Unleashes Grok 4.1, Featuring End-to-End Encrypted Chat for Enhanced Security and User Privacy
In a move that underscores its commitment to user privacy and cutting-edge AI, X has launched Grok 4.1, the latest iteration of its AI model, alongside a new end-to-end encrypted chat feature. This update marks a significant leap forward in conversational AI, boasting improvements in emotional intelligence, creative expression, and a reduction in factual inaccuracies.
Grok 4.1: A Smarter, Faster, and More Empathetic AI
xAI, the company behind Grok, has focused on enhancing the model's real-world usability. Grok 4.1 is designed to be more perceptive to nuanced intent, more engaging in conversation, and more consistent in its personality. The company has achieved this through large-scale reinforcement learning, optimizing the model's style, helpfulness, and alignment.
One of the most notable improvements is Grok 4.1's enhanced emotional intelligence. The model can now better recognize tone and emotional changes in prompts, responding with more human-like emotional logic and generating comforting and understanding replies. Internal evaluations show that Grok 4.1 outperforms models like Gemini 2.5 Pro, Claude Opus 4, and GPT-5 on the EQ-Bench benchmark. This allows Grok 4.1 to provide longer, more empathetic responses.
Grok 4.1 also excels in creative writing. It demonstrates stronger language style and imagination in writing, storytelling, and social contexts. The model scored 1708.6 on the Creative Writing v3 benchmark, outperforming Claude 4.5 Sonnet.
Furthermore, xAI has addressed the issue of "hallucinations," where AI models generate inaccurate information. Grok 4.1 has a significantly reduced hallucination rate compared to its predecessors. Internal evaluations show a reduction in error rate from 12% to 4% in information-seeking prompts.
Grok 4.1 is available in two modes: Thinking and Non-Thinking. The Thinking mode (codename Quasarflux) exposes explicit reasoning steps before providing the final answer, making it suitable for complex tasks. It holds the top position on the LMArena Text Arena leaderboard with an Elo score of 1483. The Non-Thinking mode (codename Tensor) bypasses intermediate tokens for a faster response. It ranks second on the same leaderboard with an Elo of 1465, outperforming other models' full-reasoning configurations.
End-to-End Encrypted Chat for Enhanced Privacy
In addition to the Grok 4.1 improvements, X has rolled out end-to-end encrypted chats across its platform. This feature provides users with a secure messaging environment, protecting their conversations from unauthorized access. The encrypted chat includes file sharing, the ability to edit/delete messages, screenshot blocking and detection features, and no ads or tracking.
Availability
Grok 4.1 is now accessible to all eligible users through the official Grok website, the X platform, and dedicated iOS and Android applications. The model is the default in Auto mode and can be selected manually.
