OpenAI announces new agentic coding model "GPT-5.2-Codex"
On December 18, 2025, OpenAI announced “GPT-5.2-Codex,” a new model built by optimizing GPT-5.2 for agentic coding. Its standout features are major improvements in the persistence needed to see large, long-running tasks through to completion, execution performance on Windows, and cybersecurity-related capability, and it posted top-tier results on both “SWE-Bench Pro” and “Terminal-Bench 2.0,” benchmarks that evaluate realistic software engineering tasks.
Details
- Handling long-running tasks: “Context compaction,” which compresses context while continuing to work, improves persistence on long-horizon work
- Strength on large-scale changes: Excels at wide-reaching code changes such as large refactors and migrations
- Improved Windows support: Execution performance on Windows improved significantly compared to previous models
- Cybersecurity capability: Gained “markedly stronger” cybersecurity-related capability than previous models; also intended for defensive security use through an invite-only “trusted access pilot” program for vetted experts and organizations
- Improved visual understanding: More accurate interpretation of screenshots, diagrams, and UI elements
- Benchmarks: Achieved top-tier scores on both “SWE-Bench Pro,” which evaluates realistic software development tasks, and “Terminal-Bench 2.0,” which evaluates tasks like compiling code, training models, and setting up servers in a terminal
- Availability: Available the same day across all Codex surfaces on paid ChatGPT plans. API availability was promised “within a few weeks”
What happened next
GPT-5.2-Codex became another example of Codex’s dedicated models being updated at less than six-month intervals. About a month and a half later, in February 2026, its successor “GPT-5.3-Codex” was announced alongside an even faster variant, “GPT-5.3-Codex-Spark,” further accelerating the pace of model updates.