国泰海通｜计算机：Grok-4引领AI进阶，掘金算力与垂直领域赛道

Core Viewpoint - The release of Grok 4 by xAI marks a significant advancement in AI technology, leading to increased exploration and innovation within the industry, pushing it towards a higher development stage [1][2]. Group 1: Performance and Capabilities - Grok 4 has achieved a tenfold increase in pre-training and reasoning computation capabilities compared to its predecessor, with a training scale reaching 100 times that of Grok 2 [2]. - In human-level testing, Grok 4 scored 45%, doubling the performance of the previous leading AI, Gemini 2.5pro, and has set new records in authoritative benchmark tests like GPQA and AIME25 [2]. - The multi-agent collaboration feature of Grok 4 Heavy has demonstrated exceptional performance, achieving full marks in AIME25, indicating a significant leap in reasoning capabilities beyond traditional human-designed tests [2]. Group 2: Real-World Applications - Grok 4 has shown revolutionary progress in solving real-world problems, with a doubling of response speed and halving of latency in voice functions, significantly enhancing user experience [3]. - In the Vending-Bench test, Grok 4 generated a net asset value of 4694.15, outperforming the second-place Claude Opus 4 by more than two times, validating its long-term strategic execution capabilities [3]. - The system has been utilized in the biomedical field to assist in screening millions of experimental data and generating research hypotheses, proving its effectiveness in complex cross-industry tasks [3]. Group 3: Future Developments - Despite its advancements, Grok 4 still has notable shortcomings in multi-modal capabilities, particularly in image understanding and generation, which require significant improvement [4]. - Future developments will focus on breakthroughs in video generation technology, aiming to create an AI video production closed loop through end-to-end training on the X platform [4]. - The ultimate goal is to build a super-intelligent agent that integrates deep thinking, real-time response, and multi-modal collaboration, fundamentally reshaping human-machine collaboration paradigms [4].