OpenAI Launches GPT-5.3-Codex
Interestingly, GPT-5.3-Codex played a critical role in its own development. The team used its earlier version to debug its own training, manage self-deployment, and diagnose test results and evaluations.
Topics
News
- Moltbook’s Episode Turns Out To Be Theatrics
- OpenAI Launches GPT-5.3-Codex
- Inside Al Baraka Bank Egypt’s First Digital Branch
- Dubai Chamber of Digital Economy Partners With Canva to Set Up Regional HQ
- Apple Puts Virtual Health Coach On Hold Amid Tough Competition
- Alphabet Remains Cautious On Apple Deal, Dodges Investor Question
[Image source: Krishna Prasad/MITSMR Middle East]
Days after announcing the retirement of several legacy AI models from ChatGPT on February 13, including GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini, to transition towards newer GPT-5 series models, OpenAI has unveiled GPT-5.3-Codex, a new iteration in its Codex series.
Deemed as the most capable agentic coding model to date, GPT‑5.3-Codex builds upon GPT‑5.2-Codex’s frontier coding and GPT‑5.2’s professional knowledge capabilities– 25% faster in a single model.
The upgrade enables users to take on tasks involving research, tool use, and complex execution.
“Much like a colleague, you can steer and interact with GPT‑5.3-Codex while it’s working, without losing context,” the official blog read.
Interestingly, GPT-5.3-Codex played a critical role in its own development. The team used its earlier version to debug its own training, manage self-deployment, and diagnose test results and evaluations.
“Our team was blown away by how much Codex was able to accelerate its own development,” the company stated.
With GPT‑5.3-Codex, OpenAI sets a new standard for SWE-Bench Pro and Terminal-Bench, while also delivering strong performance on OSWorld and GDPval, the four benchmarks it uses to measure coding, agentic, and real-world capabilities.
While SWE‑bench Verified only tested Python, SWE‑Bench Pro does so across four languages and is more contamination‑resistant, challenging, diverse and industry-relevant. Meanwhile, the model far exceeds the previous performance on Terminal-Bench 2.0 for measuring terminal skills, that too with fewer tokens than any prior model.
GPT‑5.3-Codex fares better on understanding intent compared to GPT‑5.2-Codex.
In web development, Codex showcased capabilities to build complex applications, such as games, through autonomy in iterative development with minimal human intervention.
Going beyond coding, software engineers, designers, product managers, and data scientists will be able to use GPT‑5.3‑Codex to support the software lifecycle, from debugging, deploying, monitoring, writing PRDs, to editing copy, user research, tests, and metrics.
Security remains a critical area of focus as OpenAI prepares for strengthened cyber safeguards to support defensive use and broader ecosystem resilience.
GPT‑5.3-Codex is the first model to be classified as high capability for cybersecurity-related tasks under the startup’s Preparedness Framework. It is also the first to be directly trained to identify software vulnerabilities.
Co-designed for, trained with, and served on NVIDIA GB200 NVL72 systems, Codex is currently accessible with paid ChatGPT plans, with API access planned for future roll-out. With this, OpenAI progresses towards shaping models as end-to-end partners for complex tasks, beyond mere tools.
