GPT-5.3-Codex: Code 25% Faster Now
Unlock 25% faster agentic coding that builds itself—revolutionize your dev workflow now.
Feb 6, 2026 (Updated Feb 16, 2026) - Written by Lorenzo Pellegrini
This image is part of OpenAI's official brand assets, available from their press kit
Lorenzo Pellegrini
Feb 6, 2026 (Updated Feb 16, 2026)
OpenAI Unveils GPT-5.3-Codex: The Most Powerful Agentic Coding Model Yet
OpenAI has launched GPT-5.3-Codex, a groundbreaking self-improving agentic coding model that redefines software engineering and professional workflows. This release combines superior coding performance with advanced reasoning, delivering 25% faster speeds and real-time interactivity for developers everywhere.
What is GPT-5.3-Codex?
GPT-5.3-Codex builds on the strengths of GPT-5.2-Codex by integrating frontier coding abilities with enhanced reasoning and professional knowledge. It excels in complex, real-world tasks, transforming Codex from a code-writing tool into a full-spectrum agent capable of nearly any computer-based professional work. The model sets new benchmarks on SWE-Bench Pro, Terminal-Bench 2.0, OSWorld, and GDPval, outperforming previous versions across coding, agentic, and real-world capabilities.
Key Features and Improvements
This model introduces several innovations that boost productivity and collaboration:
- 25% Faster Performance: Optimized inference delivers quicker results in the Codex app, CLI, IDE extensions, and web.
- Interactive Steering: Users can submit messages mid-task to guide the agent without losing context, with stable steer mode enabled by default.
- Multi-Language Excellence: Superior handling of diverse programming languages, topping benchmarks like SWE-Bench Pro.
- Terminal and OS Mastery: Enhanced accuracy for system-level interactions on Terminal-Bench 2.0 and OSWorld-Verified.
- Web and Frontend Prowess: Generates production-quality websites from simple prompts with improved UX/UI.
- Cybersecurity Advances: Direct vulnerability detection and the highest capability rating under OpenAI's Preparedness Framework.
- Broad Workflow Support: Boosts efficiency in documentation, presentations, data analysis, and structured outputs with fewer tokens used.
Self-Improving Nature: A Model That Builds Itself
One standout aspect is GPT-5.3-Codex's role in its own creation. Early versions assisted in debugging training runs, supporting deployments, diagnosing evaluations, and managing operational tasks like scaling GPU clusters. Trained and served on NVIDIA GB200 NVL72 systems, it accelerated OpenAI's research, fundamentally changing how teams work.
Availability and Access
Paid ChatGPT users can access GPT-5.3-Codex immediately across all Codex platforms: the app, CLI, IDE extensions, and web. To switch models, select it from the composer dropdown after signing in. API access is coming soon, with safeguards for high-risk uses like cybersecurity to mitigate potential harms.
Cybersecurity Considerations and Safeguards
Classified as "high capability" for cybersecurity, GPT-5.3-Codex prompts new safety measures. OpenAI delays unrestricted API access and introduces a trusted-access program for vetted professionals. These steps address risks from its advanced coding and reasoning powers, ensuring responsible deployment while enabling everyday development tasks.
Conclusion
GPT-5.3-Codex marks a pivotal advancement in AI-driven coding, empowering developers with unprecedented speed, interactivity, and versatility.
This model not only elevates software engineering but also hints at broader professional transformations powered by agentic AI. Developers should explore it now to experience the future of coding firsthand.
GPT-5.3-Codex marks a significant step forward in agentic AI for coding. At IntraMind, we value how models like this can accelerate professional workflows, while emphasizing the importance of expert human oversight to ensure quality and security in real-world projects. It's a powerful tool for developers exploring new frontiers of productivity.