Premium

GPT 5 Codex gets new updates and can now review codes

Developers can personalise security configurations to suit their risk tolerance, and authorise commands with full access or permit the agent to utilise web search and connect to the MCP server.

GPT-5-Codex is designed for use within the Codex CLI, IDE extension, Codex cloud environment, and GitHub. (Image: OpenAI)

ChatGPT has introduced improvements to the GPT-5-Codex, a version of GPT-5 specially designed for agentic coding inside Codex. This edition is tailored for actual software engineering tasks and excels in fast, interactive sessions as well as in handling long, complex projects on its own. Its code review functionality is capable of detecting severe bugs before deployment.

GPT-5-Codex is available wherever Codex is being utilised, serving as the default for cloud tasks and code review. Developers can obtain it for local tasks through the Codex CLI and the IDE extension.

style="display:none;">

OpenAI has upgraded its GPT-5-Codex model to report results on all 500 tasks, including refactoring tasks from large repositories and Python, Go, and OCaml languages. The model scales its thinking time based on task complexity, blending two core abilities of a coding agent: collaboration with developers in interactive sessions and independent work on long tasks.

Story continues below this ad

This results in Codex being perceived as faster on small, clearly defined requests and more capable of tackling lengthy tasks such as extensive refactors.

Also Read | What is Codex, OpenAI’s latest AI coding agent capable of multitasking?

In testing, GPT-5-Codex has been seen to operate independently for more than seven hours on complex, big tasks, refactoring its implementation, resolving test failures, and producing successful outcomes. On OpenAI employee traffic, GPT-5-Codex spends 93.7 per cent fewer tokens on the bottom 10 per cent of user turns when sorted by model-generated tokens and more time reasoning, revising, testing code, and refactoring on the top 10 per cent.

Codex is adept at performing code reviews and finding major flaws. It navigates your codebase, inspects dependencies, and executes tests to check for correctness. GPT-5-Codex comments are less likely to be unimportant or incorrect, allowing the user to concentrate on important issues. It serves as a reliable ally for front-end work, building beautiful desktop apps, and improving human preference evaluations for mobile websites.

GPT-5-Codex is designed for use within the Codex CLI, IDE extension, Codex cloud environment, and GitHub. The latest improvements enhance Codex as a collaborative programmer, featuring a redesigned CLI and a new IDE extension.

Story continues below this ad

The Codex CLI has been reconstituted to support agentic coding pipelines, strengthening its capabilities to enable users to add and share images, monitor work with a to-do list, and use tools like web search and MCP for integrating with outside systems. The new IDE extension and GitHub support bring the Codex cloud agent closer to developer pipelines, making task delegation possible without exiting the editor or GitHub.

Also Read | OpenAI debuts new AI coding agent that writes, edits, and executes commands locally

Codex can now offer code review capabilities to identify critical defects in GitHub repositories, checking that pull requests conform to the true diff, reasoning throughout the whole codebase and dependencies, and running code and tests to check behaviour.

Developers can customise security configurations to suit their risk tolerance, limit network access to trusted domains in the cloud, and authorise commands with full access or permit the agent to utilise web search and connect to MCP servers.

ChatGPT offers various plans, such as Codex, for different levels of coding activities. Users accessing Codex CLI via API will soon be able to utilise GPT-5-Codex through API.

Tags:
ChatGPT
Gpt 5