OpenAI has expanded Codex with GPT-5-Codex, a variant of GPT-5 that has been specifically trained for software development. This new engine makes the programming assistant faster, more reliable, and more independent, both in short sessions and in complex processes lasting several hours.
Since the launch of Codex CLI and the web version earlier this year, the tool has become a permanent programming partner. Thanks to integration with ChatGPT accounts, developers can now seamlessly switch between local and cloud environments without losing context. Codex is available in the terminal, IDEs, GitHub, and the ChatGPT app and is included in all paid ChatGPT subscriptions.
GPT-5-Codex has been trained on additional coding data and is primarily focused on realistic software engineering tasks. The model can set up projects from scratch, extend features, perform large-scale refactors, and provide code reviews. In benchmarks, it achieved a score of 51.3 percent in refactoring, more than seventeen points higher than GPT-5. During internal tests, the model worked independently for more than seven hours, detecting its own errors, performing tests, and correcting results.
The power of GPT-5-Codex lies in its flexibility. It responds faster than GPT-5 to small requests, while taking more time to reason and test complex tasks. In practice, the model uses far fewer tokens for simple tasks and more for challenging ones, resulting in more appropriate output. Developers also need fewer detailed instructions: the AI largely follows best practices and coding standards independently.
New tools
In addition to the new engine, the tools have also been updated. The open-source CLI now supports sharing images such as wireframes and UI sketches to build context. The system tracks progress through a task list and can utilize external tools, such as web searches. The interface in the terminal has been made clearer and gives developers more control over what Codex is allowed to do.
Another new feature is the IDE extension, which works in VS Code, among others. Because Codex has access to open files and selected code, prompts are shorter and results are faster. Developers can easily move work between the cloud and local environments. In the cloud environment, performance has been enhanced through caching, resulting in a 90% reduction in the average turnaround time.
Codex can also automatically review pull requests and flag critical issues. It is already being used extensively within OpenAI, and according to a tech lead at Cisco Meraki, the tool helped his team to completely outsource refactoring and test generation, enabling them to meet their release schedule without additional risk.