OpenAI GPT-5.2 模型发布，深耕表格 PPT 与超长上下文处理

Core Insights - OpenAI has officially launched its advanced AI model GPT-5.2, aimed at developers and professional users, to unlock more economic value through technological innovation [1] Product Features - GPT-5.2 will be available through API for paid ChatGPT users and developers, categorized into three types: Instant, Thinking, and Pro [3] - The Instant category focuses on efficient responses for tasks like information retrieval, content creation, and language translation [3] - The Thinking category excels in solving structured problems, particularly in coding, long document analysis, mathematical operations, and planning [3] - The Pro category targets high-difficulty professional scenarios, emphasizing the robustness and reliability of output [3] Performance Enhancements - GPT-5.2 has made significant breakthroughs in key capabilities, particularly in table creation, presentation generation, code writing, image understanding, long context processing, and multi-step toolchain task execution [3] - The model shows comprehensive performance improvements across six core areas: coding, mathematics, science, visual understanding, long text reasoning, and tool usage [3] Benchmark Testing - OpenAI's benchmark tests indicate that the Thinking category of GPT-5.2 outperforms similar models like Gemini 3 and Claude Opus-4.5 in nearly all reasoning tests, covering real software engineering cases and high-level scientific knowledge applications [4] - The enhanced mathematical capabilities of GPT-5.2 help maintain consistency in multi-step logic, reducing the accumulation of minor errors in long sequence tasks, which is beneficial for financial modeling and data processing [4] Industry Feedback - In the coding development area, GPT-5.2 has shown remarkable improvements, with feedback from startup teams indicating that its programming capabilities have reached industry-leading standards, resulting in measurable efficiency gains [4] - The error rate in the Thinking category has decreased by 38% compared to the previous model, significantly enhancing reliability in daily decision-making, academic research, and content creation [4]