OpenAI GPT-5.3-Codex Launched: New AI Agentic Model Can Autonomously Build, Debug and Manage Software
OpenAI has launched GPT-5.3-Codex, its most advanced agentic coding model to date. Featuring a 25% speed increase over its predecessor, the model can autonomously manage the entire software lifecycle, from research to deployment. It notably assisted in its own development, marking a significant milestone in self-improving artificial intelligence capabilities.
San Francisco, February 6: OpenAI has introduced GPT-5.3-Codex, a new flagship model designed to function as a general-purpose agent capable of executing complex technical work. The model integrates the reasoning power of GPT-5.2 with enhanced frontier coding performance, allowing it to handle long-running tasks that require tool use, autonomous research, and real-time interaction with human collaborators.
In a landmark shift for the company’s development process, OpenAI revealed that the Codex team utilized early versions of GPT-5.3-Codex to build the model itself. The AI was instrumental in debugging its own training runs, managing GPU cluster deployments during traffic surges, and diagnosing complex evaluation results, significantly accelerating the research and engineering timeline. OpenAI Frontier Platform Launched To Help Enterprises Deploy and Manage AI Agents As ‘Coworkers’.
Record-Breaking Performance in Software Engine ering
The new model has set industry benchmarks for coding and agentic capabilities. According to OpenAI, GPT-5.3-Codex has achieved state-of-the-art results on SWE-Bench Pro and Terminal-Bench 2.0. Unlike previous evaluations that focused solely on Python, these benchmarks test the model across multiple programming languages and terminal-based environments.
Beyond standard code generation, the model demonstrates advanced web development skills. In internal tests, it successfully built complex, functional games from scratch over several days using only generic follow-up prompts. The model also shows a refined understanding of user intent, defaulting to production-ready features such as automated carousels and logical pricing displays without explicit instructions.
Agentic Computer Use and Professional Knowledge
GPT-5.3-Codex is designed to support the entire software lifecycle, including writing product requirement documents, conducting user research, and monitoring metrics. Its capabilities extend to general professional tasks; in GDPval tests—an evaluation measuring 44 different occupations—the model matched top-tier performance in creating presentations and spreadsheets.
The model also showed a significant "step change" in visual computer use. On the OSWorld benchmark, which requires agents to complete productivity tasks within a visual desktop environment, GPT-5.3-Codex far outperformed previous iterations, moving closer to becoming a single, general-purpose collaborator for all computer-based work.
Enhanced Cybersecurity Measures and Trusted Access
As AI capabilities reach new heights, OpenAI has classified GPT-5.3-Codex as "High capability" for cybersecurity under its Preparedness Framework. It is the first model directly trained to identify software vulnerabilities. To mitigate potential misuse, the company is deploying its most comprehensive safety stack, including automated monitoring and threat intelligence. OpenAI Launches Codex App for macOS, Brings Multi-Agent Control and Parallel AI Workflows; Windows Support Coming Soon.
To bolster defensive research, OpenAI is launching "Trusted Access for Cyber," a pilot program for security professionals. Additionally, the company has committed USD 10,000,000 in API credits to accelerate cyber defence for open-source software and critical infrastructure. The model is currently available to paid ChatGPT users via the Codex app, CLI, and IDE extensions.
(The above story first appeared on LatestLY on Feb 06, 2026 07:24 AM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).