强化AI编程能力迎战谷歌！OpenAI发布GPT-5.2-Codex，软件工程和网安一把抓

Core Insights - OpenAI has launched the GPT-5.2-Codex model, an advanced coding model focusing on software engineering and cybersecurity, enhancing its competitive edge against Google's Gemini [1][2] - The new model shows significant improvements in coding performance, cybersecurity capabilities, and long-term task handling, achieving record accuracy in benchmark tests [1][3] Group 1: Model Performance - GPT-5.2-Codex achieved an accuracy of 56.4% in the SWE-Bench Pro test, surpassing GPT-5.2's 55.6% and GPT-5.1's 50.8% [3] - In the Terminal-Bench 2.0 test, GPT-5.2-Codex reached an accuracy of 64.0%, compared to GPT-5.2's 62.2% and GPT-5.1's 58.1% [3] - The model has been optimized for long-term coding tasks, improving context retention and reliability in large-scale projects [5][8] Group 2: Cybersecurity Enhancements - GPT-5.2-Codex has made significant strides in cybersecurity, with capabilities observed to have dramatically improved since the previous versions [8] - The model has demonstrated the ability to tackle advanced multi-step challenges requiring professional-level cybersecurity skills [8] - A real-world case highlighted the model's potential in defensive cybersecurity, where vulnerabilities were discovered and responsibly disclosed using the previous Codex model [9] Group 3: Access and Security Measures - OpenAI is implementing additional protective measures for the enhanced cybersecurity capabilities, including specialized training and a trusted access program for security professionals [11] - The trusted access program is initially available to vetted security professionals and organizations with specific cybersecurity use cases, allowing them to utilize OpenAI's models for defensive work [11]