Core Insights - The latest AI model Grok 4.1 from xAI, led by Elon Musk, has been released, claiming to compete directly with GPT-5.1 [2] - Grok 4.1 has two versions: Grok 4.1 Thinking (reasoning) and Grok 4.1 (non-reasoning), both available for free on various platforms [2] - Grok 4.1 Thinking achieved a score of 1483 in the LMArena ranking, surpassing Gemini 2.5 Pro by 31 points, while the non-reasoning version ranked second [2][3] Performance Metrics - Grok 4.1 Thinking scored 1586 in the EQ-Bench3 emotional intelligence test, also ranking first, with the non-reasoning version in second place [3] - Compared to its predecessor, Grok 4.1 shows significant improvements in empathy and creative writing, with a score increase of 600 points and a reduction in hallucination rate from 12.09% to 4.22% [4][5] Development Insights - The enhancements in Grok 4.1 are attributed to advancements in reinforcement learning, with the scale of reinforcement learning expanded tenfold [7] - xAI's team composition has shifted to a predominantly Chinese workforce, which is part of Musk's talent strategy [7] Future Outlook - The anticipated Grok 5 model has been delayed due to resource constraints and rigorous testing requirements, with an expected release next year [9] - Grok 5 is projected to have around 60 trillion parameters, aiming to be the most intelligent AI, which necessitates extensive testing and optimization [9]
马斯克亮AI王牌：Grok 4.1发布，智商情商双在线霸榜，背后团队被传是“全华班”
Sou Hu Cai Jing·2025-11-18 06:54