Grok 4正式发布！当世最强算力训练，对垒GPT-5和Claude 4 Opus

Core Insights - Grok 4 is the latest flagship model from xAI, featuring a context window of 256,000 tokens, enhanced reasoning capabilities, and a multi-modal function that supports complex interactions [1][3][5] - The subscription fee for Grok 4 is set at $30 per month, while the Heavy version costs $300 per month, with Grok 3 remaining free [3][4] - Grok 4 has been trained on xAI's Colossus supercomputer, boasting reasoning abilities that surpass those of most graduate students [3][5] Performance and Features - Grok 4's reasoning capability has improved tenfold compared to its predecessor, and it is expected to perform comparably to GPT-5 and Claude 4 Opus [5][6] - In benchmark tests against models like GPT-3, Gemini 2.5 Pro, and Claude 4 Opus, Grok 4 consistently scored among the top [7] - The model achieved a record accuracy of 26.9% in the Humanity's Last Exam benchmark test, indicating strong autonomous reasoning [9] Development Timeline - The Grok 4 API is now available, with a programming version set to launch in August, a multi-modal agent version in September, and a video model in October [4] Market Position and Controversies - Grok 4 includes a DeepSearch feature that allows it to extract real-time data from the internet, enhancing its understanding of internet culture [12] - The release of Grok 4 coincides with a critical transition period for xAI, following its merger with X in March [12] - The model has faced controversy over its "free speech" approach, raising questions about its suitability for human interaction [13]