GPT-5.3-Codex represents a groundbreaking leap in artificial intelligence as OpenAI introduces the most capable agentic coding model to date. This revolutionary AI system combines frontier coding performance with advanced reasoning capabilities, marking a significant milestone in the evolution of AI-assisted software development.
Released on February 2, 2026, the new model advances both the coding performance of its predecessor GPT-5.2-Codex and the reasoning capabilities of GPT-5.2, delivering a unified solution that operates 25% faster. What makes this release particularly remarkable is that GPT-5.3-Codex was instrumental in creating itself, with the Codex team using early versions to debug training, manage deployment, and diagnose test results.
Revolutionary Self-Development Capabilities
The model represents a paradigm shift in AI development methodology. For the first time, an AI system actively participated in its own creation and refinement. The Codex team leveraged early versions of GPT-5.3-Codex to accelerate development cycles, with team members reporting that the AI significantly expedited debugging infrastructure, tracked training patterns, and built rich applications for human researchers to understand behavioral differences from prior models.
Engineering teams used the system to optimize harnesses, identify context rendering bugs, and root cause low cache hit rates. During deployment, the model dynamically scaled GPU clusters to adjust to traffic surges while maintaining stable latency, demonstrating unprecedented autonomous operational capabilities.
Industry-Leading Benchmark Performance
GPT-5.3-Codex sets new industry standards across multiple evaluation frameworks. The model achieved state-of-the-art performance on SWE-Bench Pro, scoring 77.3% accuracy compared to 64.0% for GPT-5.2-Codex and 62.2% for GPT-5.2. This rigorous evaluation tests real-world software engineering across four programming languages, making it more contamination-resistant and industry-relevant than previous benchmarks.
On OSWorld-Verified, which measures computer-use capabilities in visual desktop environments, the model demonstrated 64.7% accuracy compared to 38.2% for its predecessor. Remarkably, it achieves these results using fewer tokens than any prior model, allowing users to build more with less computational overhead.
Beyond Code Generation: Complete Professional Workflows
The system transcends traditional code generation, supporting the entire software development lifecycle including debugging, deploying, monitoring, writing product requirement documents, editing copy, user research, tests, and metrics analysis. Its agentic capabilities extend beyond software engineering to knowledge work across 44 occupations, matching GPT-5.2 performance on GDPval evaluations.
Demonstrations include autonomous development of complex games over millions of tokens, including a sophisticated diving game featuring reef exploration, fish collection mechanics, and resource management systems. The model also excels at creating production-ready landing pages with sensible defaults, automatically implementing features like discounted pricing displays and transitioning testimonial carousels.
Enhanced Security and Responsible Deployment
OpenAI classifies GPT-5.3-Codex as “High capability” for cybersecurity-related tasks under its Preparedness Framework, making it the first model directly trained to identify software vulnerabilities. The company deployed comprehensive safeguards including safety training, automated monitoring, trusted access programs, and enforcement pipelines.
To accelerate defensive capabilities while preventing misuse, OpenAI launched the Trusted Access for Cyber pilot program and committed $10 million in API credits through its Cybersecurity Grant Program. The company is expanding the private beta of Aardvark, its security research agent, and partnering with open-source maintainers to provide free codebase scanning for widely-used projects like Next.js.
Interactive Collaboration and Real-Time Steering
The model introduces enhanced interactivity, providing frequent updates on key decisions and progress during task execution. Users can interact in real-time, asking questions, discussing approaches, and steering solutions without losing context. This collaborative approach transforms the AI from a tool into a colleague that maintains continuous dialogue throughout complex, long-running tasks.
GPT-5.3-Codex is available with paid ChatGPT plans across all Codex platforms including the app, CLI, IDE extension, and web interface. The model runs 25% faster thanks to infrastructure improvements and was co-designed for, trained with, and served on NVIDIA GB200 NVL72 systems. API access is planned for safe enablement in the near future.
