Following multiple rumors from earlier today, Anthropic finally released Claude Opus 4.6. OpenAI, being a fierce rival, has followed up the launch with its GPT‑5.3‑Codex release. The launch came moments after Anthropic dropped its model. If you haven’t checked out Opus 4.6, make sure to read our news piece here.
GPT‑5.3‑Codex isn’t just faster, it’s more collaborative, too
GPT‑5.3‑Codex is OpenAI’s latest bet towards agentic coding, which combines the coding prowess of GPT‑5.2‑Codex with the reasoning and professional knowledge of GPT‑5.2. It’s also 25% faster, meaning allows it to handle long-running tasks like research, tool use, and complex execution. Much like a colleague, you can interact with it while it works, without losing context.
One interesting thing OpenAI shared in the announcement is that GPT‑5.3‑Codex helped in creating itself. Apparently, the internal team was “blown away” by how fast it accelerated its own development. That’s not all; early versions apparently helped debug training, manage deployment, and analyze evaluations. Do note that GPT‑5.3‑Codex goes beyond writing code. In other words, it can now do nearly everything a developer or professional might do on a computer.
On SWE-Bench Pro and Terminal-Bench 2.0, GPT‑5.3‑Codex sets a new benchmark for coding performance. SWE-Bench Pro covers four programming languages and industry-level tasks, while Terminal-Bench 2.0 measures terminal-based coding skills. It achieves state-of-the-art results with fewer tokens, letting users build more efficiently.
For web development, the model can autonomously iterate over games and apps. Using preselected prompts like “fix the bug” or “improve the game,” GPT‑5.3‑Codex developed a racing game with eight maps and a diving game exploring reefs, oxygen management, and hazards. Simple website prompts now default to more functional and production-ready results, like landing pages with price toggles, testimonial carousels, and structured sections.
You may also like: Sam Altman Slams Anthropic’s Super Bowl Ad, Says the Campaign is “Clearly Dishonest”
The new model excels beyond coding
GPT‑5.3‑Codex supports the full software lifecycle: debugging, deploying, writing PRDs, analyzing metrics, and generating documents or slides. It performs strongly on GDPval, a professional knowledge benchmark, and OSWorld, which evaluates computer-based tasks visually. Its agentic abilities extend to spreadsheets, presentations, and data analysis, making it a versatile collaborator.
OpenAI has classified GPT‑5.3‑Codex as being highly capable for cybersecurity tasks. It helps find and patch software vulnerabilities while operating under strict safeguards. Codex also provides real-time updates during tasks, allowing users to steer, interact, and track progress from start to the very end.
GPT‑5.3‑Codex is now available across ChatGPT plans, the Codex app, CLI, IDE extensions, and web. What do you make of this new model? Let us know in the comments below.
