OpenAI Releases GPT-5.2-Codex for Autonomous Software Engineering
The new model features "context compaction" and enhanced cybersecurity capabilities for long-horizon tasks.
OpenAI expanded its product line on Thursday with the release of GPT-5.2-Codex, a specialized model engineered for complex, autonomous software development. Departing from general-purpose chat interactions, the model is designed to function as an "agent," capable of executing long-horizon tasks such as refactoring large codebases or conducting defensive security operations without constant human supervision.
Context Compaction Technology
A primary technical innovation in GPT-5.2-Codex is "context compaction." This architecture allows the model to compress and retain the semantic essence of extended work sessions, overcoming the memory limits that often cause performance degradation in large language models. OpenAI claims this feature …
Archive Access
This article is older than 24 hours. Create a free account to access our 7-day archive.