OpenAI releases GPT-5.2-Codex, a coding-focused AI model

On December 18, 2025 local time, OpenAI announced GPT-5.2-Codex , the most advanced agent-based coding model to date. GPT-5.2-Codex is a version of

Meet GPT-5.2-Codex, the best agentic coding model yet for complex, real-world software engineering.

With native compaction, better long-context understanding, and improved tool-calling, it is a more dependable partner for your hardest tasks.

Available in Codex starting today.…

— OpenAI Developers (@OpenAIDevs) December 18, 2025

GPT-5.2-Codex builds on the strengths of GPT-5.2 and the cutting-edge agent-driven coding and terminal operation capabilities of GPT-5.1-Codex-Max. It offers enhanced support for long-term work through context compression, improved performance for large-scale code changes such as refactoring and migration, improved performance in Windows environments, and significantly enhanced cybersecurity features. As a result, it can be used as a more reliable partner for long-term coding work while maintaining token efficiency for inference.

Enhanced vision performance enables GPT-5.2-Codex to more accurately interpret screenshots, technical diagrams, charts, and UI screens shared during coding, and builds on the capabilities introduced in GPT-5.1-Codex-Max, making agent-assisted coding in native Windows environments more effective and reliable.

GPT-5.2-Codex achieves state-of-the-art performance on SWE-Bench Pro and Terminal-Bench 2.0, benchmarks that evaluate agent performance across a variety of tasks in realistic terminal environments. These improvements enable better performance across real-world software engineering tasks, from repository navigation and refactoring to creating and reviewing pull requests.

OpenAI has demonstrated its capabilities in specialized areas such as cybersecurity, revealing that security researchers using Codex CLI and GPT-5.1-Codex-Max discovered and responsibly disclosed three vulnerabilities in React that could have led to denial of service or source code exposure.

While GPT-5.2-Codex has stronger cybersecurity capabilities than any model released by OpenAI to date and has the potential to contribute to strengthening cybersecurity at scale, it also creates new dual-use risks that require careful implementation. One reason for this is that GPT-5.2-Codex does not meet the high level of cybersecurity capabilities set by OpenAI's Preparedness Framework .

Taking into account both the capabilities and risks of GPT-5.2-Codex, OpenAI will begin offering GPT-5.2-Codex to all paid ChatGPT users on December 19th via Codex CLI, IDE extensions, cloud, and code review. At the same time, OpenAI is preparing to securely enable third-party provisioning via API. OpenAI will also launch an invitation-only pilot program to provide vetted security professionals and organizations with controlled, purpose-specific access to future advanced AI cyber capabilities. This will enable authorized defensive security operations while maintaining strong safety measures.