马斯克推最强Grok 4！人类终极测试干翻OpenAI，包月费超2千元

Group 1 - Grok 4 achieved a 25.4% accuracy rate in the "Humanity's Last Exam," surpassing Google's Gemini 2.5 Pro at 21.6% and OpenAI's o3 at 21% without tools [1][8] - With tools, Grok 4 Heavy scored 44.4%, outperforming Gemini 2.5 Pro's 26.9% [1][5] - Elon Musk stated that Grok 4's academic capabilities exceed those of PhD-level individuals across various disciplines [3][8] Group 2 - xAI launched the Super Grok Heavy subscription plan at $300 per month, allowing early access to Grok 4 Heavy and upcoming features [5][7] - Grok 4's training volume is 100 times that of Grok 2, significantly enhancing its intelligence [12] - The model's ability to interact with humanoid robots is planned for future development [17] Group 3 - Grok 4 demonstrated superior performance in multiple assessments, including GPQA and ARC-AGI-2, achieving a score of 16.2% in the latter [9][10] - The model's reasoning capabilities are expected to improve further with the integration of advanced tools for physical simulations [17][36] - xAI's enterprise division has begun utilizing Grok 4 through its API, showing impressive results in various applications [23][30] Group 4 - Grok 4's multi-modal understanding is being improved, with updates reducing voice response latency by half [21] - Future developments include a coding model and video generation capabilities, with training for a new video model starting soon [36][37] - The company aims to position Grok 4 as a strong competitor against OpenAI's upcoming GPT-5 [36]