刚刚，OpenAI正式发布o3-pro！奥特曼激动更新博客：温和的奇点

Core Insights - OpenAI has launched a new reasoning model called o3-pro, which is available for Pro subscription users and has shown significant performance improvements over its predecessor, o3 [6][7][13] - The release of o3-pro was notably different from previous launches, with a simple announcement via tweets rather than a live event [7] - OpenAI CEO Sam Altman expressed high confidence in o3-pro's capabilities, highlighting its superior performance in various tasks [7][13] Performance Evaluation - In expert evaluations, o3-pro demonstrated a clear advantage over o3, particularly in fields such as science, education, programming, data analysis, and writing [13][14] - The model was assessed using a "4/4 reliability" standard, indicating that it must answer correctly in four attempts to be considered successful [17] - Despite its higher operational costs, o3-pro's performance on the ARC-AGI evaluation dataset was comparable to o3 [19] Features and Capabilities - o3-pro supports both text and image inputs, with a context window size of 200k and a maximum output token count of 100k [22] - The model integrates various tools already available in ChatGPT, such as web search and Python execution [21] - Users must purchase a $200 Pro subscription to access o3-pro, while API pricing is set at $20 per million input tokens and $80 per million output tokens [24][28] Pricing Adjustments - Following the launch of o3-pro, the pricing for o3 has been reduced by 80%, making it significantly cheaper at $2 per million input tokens and $8 per million output tokens [31][33] User Feedback - Early user tests have shown positive results, with users reporting faster and more accurate responses compared to previous models [39][45] - Some users expressed disappointment, indicating that not all expectations were met [51] Future Outlook - Sam Altman's blog post, "The Gentle Singularity," discusses the potential of AI to significantly enhance productivity and scientific progress, suggesting a future where AI systems surpass human capabilities in various domains [55][58] - The blog emphasizes the importance of ensuring that AI remains accessible and beneficial to society, advocating for a collaborative approach to AI development [68][69]