OpenAI released GPT-5.2-Codex on December 18, 2025, solidifying its status as the leading agentic coding model for software engineering and cybersecurity tasks. Enhancements include improved context compaction for long sessions, better handling of code migrations, and advanced capabilities in cybersecurity. Following intensified competition from Google’s Gemini 3, this model marks a significant shift toward specialized AI agents capable of addressing complex vulnerabilities in extensive codebases.
With an impressive accuracy of 56.4% on SWE-Bench Pro and 64.0% on Terminal-Bench 2.0, GPT-5.2-Codex sets new benchmarks for coding efficiency. Its upgraded features assist in real-world applications, notably uncovering critical vulnerabilities in software systems. OpenAI plans a controlled rollout, ensuring ethical use through a trusted access pilot for cybersecurity experts.
This launch highlights AI’s pivotal role in modern software development, offering security enhancements while navigating the dual-use risks associated with advanced technology.
Source link
