OpenAI has significantly upgraded its Codex platform, integrating a specialized version of GPT-5, dubbed GPT-5-Codex, to deliver a more autonomous and reliable coding assistant for developers. This advancement, announced on September 15, 2025, aims to transform how software engineering tasks are approached, from interactive coding sessions to complex, independent project execution. [1]

Accelerating development with GPT-5-Codex

The core of the recent upgrades is GPT-5-Codex, a version of OpenAI's GPT-5 model specifically fine-tuned for agentic software engineering workflows. This model has been trained on a comprehensive array of real-world engineering tasks, encompassing full project builds, feature additions, debugging, large-scale refactors, and code reviews. [1] Its design emphasizes improved steerability and adherence to instructions, producing higher-quality code without requiring extensive style or cleanliness directives. [1]

What is Codex?

Codex is an AI system developed by OpenAI that translates natural language into code, enabling developers to write software more efficiently. The platform supports a range of programming tasks and integrates with common development tools. [1]

Internally, GPT-5-Codex dynamically adjusts its processing time based on task complexity. For instance, it uses significantly fewer tokens for smaller, interactive requests, resulting in snappier responses. Conversely, for demanding tasks like major refactoring efforts, it dedicates more time to reasoning, editing, and testing, demonstrating independent execution for over seven hours during internal testing. [1] This adaptive approach allows it to function effectively both as an interactive pair programmer and an autonomous agent for prolonged, complex assignments. [1]

Enhanced code review capabilities

A notable improvement in GPT-5-Codex lies in its code review functionality. The model is specifically trained to identify critical flaws by navigating codebases, reasoning through dependencies, and executing code and tests to validate correctness. [1] This contrasts with static analysis tools by matching the stated intent of a pull request (PR) against the actual changes. [1] In evaluations, GPT-5-Codex generated fewer incorrect or unimportant comments compared to GPT-5, allowing human reviewers to focus on high-impact issues. [1] When enabled for GitHub repositories, Codex can automatically review PRs, posting analyses and even implementing suggested edits within the same thread. [1] OpenAI's internal use shows Codex reviewing the vast majority of their PRs, catching hundreds of issues daily. [1]

Seamless integration across developer workflows

OpenAI has unified Codex into a single product experience accessible via ChatGPT accounts, enabling developers to transition tasks between local environments and the cloud without losing context. [1] Codex now operates across various developer touchpoints, including:

• Codex CLI: The rebuilt command-line interface now supports agentic coding workflows, allowing users to attach and share images like screenshots and wireframes for shared context. [1] It also includes a to-do list for tracking progress and tools like web search for connecting to external systems. [1]

• Codex IDE extension: This extension integrates the Codex agent directly into VS Code, Cursor, and other VS Code forks. [1] It provides faster results by leveraging local context, such as open files or selected code, and facilitates moving work between cloud and local environments. [1]

• Codex cloud: Performance improvements include a 90% reduction in median completion time for new and follow-up tasks due to container caching. [1] The cloud environment automatically sets up its own environment, executes setup scripts, and can fetch dependencies at runtime with configurable internet access. [1] It also visually inspects progress and provides screenshots of its work. [1]

• GitHub integration: Developers can delegate tasks and utilize the code review feature directly within GitHub, maintaining a streamlined workflow. [1]

To learn more about the Codex features, reference OpenAI’s Developer resources

Building safe and trustworthy AI agents

OpenAI prioritizes protecting code and data while guarding against misuse. By default, Codex runs in a sandboxed environment with network access disabled, mitigating risks of harmful actions or prompt injections. [1] Developers can customize security settings to align with their risk tolerance, such as limiting network access to trusted domains in the cloud or approving commands with full access in CLI and IDE extensions. [1] OpenAI encourages developers to review the agent's work, with Codex providing citations, terminal logs, and test results for verification. [1] Consistent with GPT-5's risk assessment, GPT-5-Codex is treated as having "High capability" in biological and chemical domains, with safeguards in place. [1]

Why this matters

These upgrades mean developers can now offload more routine or complex coding tasks to an AI assistant, freeing up time for higher-level problem-solving and innovation. The improved accuracy in refactoring and code review can lead to fewer bugs and a more robust codebase, directly impacting product quality and development cycles. For organizations looking to integrate GPT-5-Codex into existing workflows, these updates can significantly enhance team productivity, reduce technical debt, and accelerate project delivery. The unified experience across development environments and enhanced safety features provide a powerful yet controlled tool for maintaining a competitive edge in software development.

What to watch

Organizations should monitor the real-world performance of GPT-5-Codex in diverse development scenarios. Particular attention should be paid to how its "agentic" capabilities evolve, especially regarding autonomous task completion and complex problem-solving. Evaluating its impact on developer efficiency and code quality within specific teams will be crucial for maximizing its strategic value.

Pricing and availability

Codex is included with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans, with usage scaling based on the subscription tier. [1] Business plans offer credit purchases for extended usage, while Enterprise plans provide a shared credit pool. [1] GPT-5-Codex will also be made available via API key for developers using Codex CLI soon. [1]