New large language models seem to spring up daily, and OpenAI’s (OPENAI) latest, GPT-5.3-Codex, was the startup’s first model that was instrumental in building itself.
“The Codex team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations—our team was blown away by how much Codex was able to accelerate its own development,” OpenAI said.
The Microsoft-backed (MSFT) company said the new model advances on the coding performance of GPT-5.2-Codex and GPT-5.2’s reasoning capabilities while producing output at a 25% faster rate.
“With GPT‑5.3-Codex, Codex goes from an agent that can write and review code to an agent that can do nearly anything developers and professionals can do on a computer,” OpenAI said.
GPT-5.3-Codex also requires about half the amount of tokens of prior models when producing certain outputs. Its accuracy rating on the Terminal-Bench 2.0 benchmark increased to 77.3% compared to 64% on GPT-5.2-Codex. It also demonstrates improvement in web and game development.
It is currently available for paid subscribers of ChatGPT, and it will be available soon on the API.
The new model was trained on Nvidia’s (NVDA) GB200 NVL72 systems.
“We are grateful to NVIDIA for their partnership,” OpenAI said.
The model was released the same day OpenAI announced its Frontier platform for enterprises.
Rival Anthropic (ANTHRO) released Claude Opus 4.6 today as well. Anthropic noted that it outperformed Google’s (GOOG)(GOOGL) Gemini 3 Pro and OpenAI’s GPT-5.2 across a range of benchmarks.