OpenAI Releases GPT-5.3-Codex, a New Codex Model for Agent-Style Development
On February 5th, 2026, OpenAI announced the release of GPT-5.3-Codex, an advanced Codex model designed specifically for agent-style development workflows. This new model enhances the capabilities of developers by allowing it to utilize tools, operate a computer, and complete longer tasks from start to finish. OpenAI claims that GPT-5.3-Codex operates 25% faster for Codex users and is accessible to all paid ChatGPT plans across various platforms including the Codex app, Command Line Interface (CLI), Integrated Development Environment (IDE) extensions, and web applications. API access is expected to follow once safety measures are fully implemented.
Key Features of GPT-5.3-Codex
The introduction of GPT-5.3-Codex brings a host of new features aimed at improving the coding experience for developers. Below are some of the standout features:
- Agentic Coding Model: The model is specifically designed for longer, tool-using workflows, making it more suitable for complex coding tasks.
- Faster Interactions: OpenAI reports that users will experience 25% faster interactions when using the Codex.
- Improved Performance: The model shows stronger performance on coding and computer-use benchmarks, making it a reliable tool for developers.
- Interactive Supervision: The Codex app now offers more interactive supervision, allowing users to ask questions and adjust their approach mid-task.
- Self-Debugging Capabilities: Early versions of the model were utilized to debug its own training, demonstrating its advanced operational capabilities.
- Cybersecurity Classification: GPT-5.3-Codex is the first OpenAI model classified as “High capability” for cybersecurity tasks, enhancing its utility in secure environments.
Performance Benchmarks
OpenAI has provided performance benchmarks that highlight the capabilities of GPT-5.3-Codex compared to its predecessors. The following benchmarks were reported:
| Benchmark | GPT-5.3-Codex | GPT-5.2-Codex | GPT-5.2 |
|---|---|---|---|
| SWE-Bench Pro (Public) | 56.8% | 56.4% | 55.6% |
| Terminal-Bench 2.0 | 77.3% | 64.0% | 62.2% |
| OSWorld-Verified | 64.7% | 38.2% | 37.9% |
| GDPval (wins or ties) | 70.9% | – | 70.9% (high) |
| Cybersecurity Capture The Flag | 77.6% | 67.4% | 67.7% |
These benchmarks demonstrate a significant improvement in performance, particularly in areas critical to software engineering and cybersecurity tasks.
Interactive Supervision in the Codex App
One of the most notable features of GPT-5.3-Codex is its enhanced interactive supervision within the Codex app. This feature allows developers to engage with the model more dynamically:
- Developers can ask questions at any point during the coding process.
- Users can discuss their approach and receive real-time feedback.
- The model can adjust its direction based on user input, leading to a more collaborative coding experience.
This level of interactivity is designed to improve the overall development process, making it easier for developers to troubleshoot issues and refine their code.
Self-Debugging and Operational Support
Another groundbreaking aspect of GPT-5.3-Codex is its ability to assist in its own training and deployment. OpenAI has revealed that early versions of the model were used to:
- Debug training runs.
- Support deployment processes.
- Diagnose evaluation results.
- Assist with operational tasks such as adapting harnesses and scaling GPU clusters based on traffic changes.
This self-supporting capability marks a significant advancement in AI technology, showcasing the potential for models to enhance their own operational efficiency.
Cybersecurity Capabilities
OpenAI has classified GPT-5.3-Codex as “High capability” for cybersecurity-related tasks under its Preparedness Framework. This classification indicates that the model is equipped to handle complex cybersecurity challenges effectively. Some key points include:
- Enhanced security measures and access controls have been implemented to safeguard sensitive tasks.
- A new pilot program, “Trusted Access for Cyber,” has been launched to further enhance cybersecurity capabilities.
These advancements make GPT-5.3-Codex a valuable tool for organizations looking to bolster their cybersecurity posture.
Availability and Infrastructure
GPT-5.3-Codex is currently available to users with paid ChatGPT plans across various platforms, including:
- Codex app
- Command Line Interface (CLI)
- Integrated Development Environment (IDE) extensions
- Web applications
OpenAI has stated that API access will be made available once it is deemed safe to do so. The model was co-designed for, trained with, and served on NVIDIA GB200 NVL72 systems, ensuring high performance and reliability.
Upgrade Notes
OpenAI has confirmed that GPT-5.3-Codex is now available in ChatGPT’s Codex surfaces. Users who rely on API availability should stay updated with OpenAI’s platform announcements to ensure they can integrate the new model into their workflows effectively.
Frequently Asked Questions
GPT-5.3-Codex features a 25% increase in interaction speed, enhanced performance on coding benchmarks, improved interactive supervision, and self-debugging capabilities, making it a more powerful tool for developers.
GPT-5.3-Codex is classified as “High capability” for cybersecurity tasks, featuring enhanced security measures and a new pilot program called “Trusted Access for Cyber” to improve its effectiveness in handling cybersecurity challenges.
API access for GPT-5.3-Codex is planned to be rolled out once it is safely enabled, and users are encouraged to keep an eye on OpenAI’s platform updates for further announcements.
Call To Action
Stay ahead in the rapidly evolving tech landscape by integrating GPT-5.3-Codex into your development processes. Explore the new features and enhance your coding efficiency today!
Note: The release of GPT-5.3-Codex represents a significant leap forward in AI-assisted coding, providing developers with powerful tools to streamline their workflows and enhance productivity.
