像第一代视网膜iPhone、堪比博士，OpenAI发布最强模型GPT-5

Core Insights - OpenAI has officially launched GPT-5, its most powerful AI system to date, available to all 700 million ChatGPT users, showcasing significant advancements in various fields such as programming, mathematics, writing, healthcare, and visual perception [1][2] Model Structure - GPT-5 consists of three key components: a base model for general questions, a deep reasoning model (GPT 5 Thinking) for complex issues, and a real-time router that optimizes model selection based on conversation type and user instructions [2] - The router continuously learns from real feedback, including user model-switching behavior and response accuracy, to enhance performance [2] Programming Capabilities - GPT-5 is described as the strongest programming model, excelling in generating and debugging large codebases, capable of creating aesthetically pleasing and responsive websites, applications, and games from a single prompt [3][4] - In benchmark tests, GPT-5 outperformed all other models, achieving a score of 74.9% in SWE-bench Verified and 88% in Aider Polyglot [4] Multimodal Abilities - The model has improved multimodal capabilities, performing well in visual, video, spatial, and scientific reasoning tasks, allowing for precise interpretation of images and non-text inputs [5][6] Safety Improvements - OpenAI conducted over 5,000 hours of safety risk testing for GPT-5, focusing on reducing instances of "hallucinations" and ensuring the model does not mislead users [7] - GPT-5 is designed to provide "safe completions" for potentially harmful queries, addressing the challenge of responding appropriately to ambiguous questions [7] Availability - GPT-5 will be available to all free and paid ChatGPT subscribers starting Thursday, with educational and enterprise customers expected to gain access the following week [7]