OpenAI has launched GPT-5-Codex, a specialized version of GPT-5 designed for advanced software engineering tasks like large-scale refactoring and code reviews. Integrated into Codex’s CLI, IDE extension, and cloud environment, this model can operate autonomously for over seven hours, generating efficient coding solutions without human input. Key features include adaptive reasoning, which tailors processing time based on task complexity, enabling collaboration on smaller tasks or persistent execution for extensive refactoring.
Performance evaluations show significant efficiency gains, with the model using 93.7% fewer tokens on simple requests compared to GPT-5, while spending additional time on complex tasks to enhance coding accuracy. GPT-5-Codex accurately handled multi-step modifications, achieving 51.3% accuracy against GPT-5’s 33.9%. Additionally, the model strengthens code reviews by analyzing dependencies and validating correctness. Early user feedback highlights its effectiveness in identifying hard bugs and delivering high-quality code. Access is available via Codex CLI and IDE extensions, with API key integration anticipated soon.
Source link
