OpenAI launches GPT-5.2-Codex with enhanced coding, cybersecurity, and vision capabilities

OpenAI has launched GPT-5.2-Codex, its most advanced model yet for complex, real-world software engineering tasks. Building on last week’s release of GPT-5.2, this version is engineered specifically for agentic coding, with optimizations that enhance its ability to handle long-horizon work through improved context compaction and stronger performance on large code changes like major refactors and migrations. It also delivers improved outcomes within Windows environments.

Alongside these core improvements, GPT-5.2-Codex features its strongest cybersecurity capabilities to date. These enhancements promise to bolster security at scale, though OpenAI notes that they introduce new dual-use risks requiring careful management. The model is better at understanding long contexts, performing reliable tool calls, and maintaining improved factuality, making it more dependable for extended coding projects while remaining efficient in its token use.

These technical advances allow Codex to work across large repositories over prolonged sessions, maintain consistent context, and complete major codebase operations. Following these upgrades, vision capabilities in GPT-5.2-Codex now support more accurate interpretation of screenshots, technical diagrams, and user interface surfaces, supporting development from design mocks to production-ready prototypes. GPT-5.2-Codex is available on all Codex surfaces for paid ChatGPT users, with API access planned in the coming weeks.