GPT-5.2-Codex

Model Overview

GPT-5.2-Codex is a specialized variant of the GPT-5.2 family, released on December 18, 2025, specifically engineered for agentic coding and software engineering. It was designed to act as an autonomous coding assistant capable of long-horizon tasks, defensive cybersecurity, and complex codebase transformations. Note: As of mid-2026, GPT-5.2-Codex is considered a legacy model, having been succeeded by newer coding models such as the GPT-5.6 series.

Capabilities

• Agentic Coding: Optimized for complex, long-running tasks like large-scale refactors and code migrations.
• Long-Context Reliability: Features advanced context compaction, maintaining consistency and project state over extended development sessions.
• Native Environment Integration: Strong performance in native OS environments (such as Windows) and highly reliable tool-calling capabilities.
• Defensive Cybersecurity: Enhanced features for vulnerability discovery, code auditing, and secure software development.
• Multimodal Translation: Capable of interpreting UI screenshots, technical diagrams, and mockups, translating them into functional code prototypes.

Example Use Cases

• Large-Scale Refactoring: Autonomously migrating legacy codebases to modern frameworks over long-running sessions.
• Security Auditing: Scanning large repositories for vulnerabilities and generating secure patches.
• UI to Code: Ingesting design mockups and outputting accurate, styled frontend code (e.g., React, Vue, HTML/CSS).
• Agentic Development: Operating as an autonomous agent to build entire features from natural language specifications.

Performance & Benchmarks

Upon its release, GPT-5.2-Codex set new state-of-the-art records for coding benchmarks:

• SWE-Bench Pro: Achieved an unprecedented 56.4% accuracy.
• Terminal-Bench 2.0: Scored 64.0% accuracy, demonstrating superior command-line and environment manipulation skills.

Intended Use & Limitations

Intended for professional software engineers, security researchers, and development teams needing an autonomous AI pair programmer. Its main limitations included the computational cost associated with long-horizon agentic tasks and potential latency. As a legacy model, its access has been phased out, and users are encouraged to adopt newer iterations for state-of-the-art coding performance.

About OpenAI

OpenAI is an AI research and deployment company dedicated to ensuring that artificial general intelligence benefits all of humanity. Known for its pioneering work on the GPT series, OpenAI continues to push the boundaries of machine learning, natural language processing, and multimodal AI systems.