On Monday, OpenAI launched Codex, an agent coding tool sold to software developers. Today, OpenAI also announced a new model designed to turbocharge Codex: GPT-5.3 Codex.
The company says this model transforms Codex from an agent that simply “writes and reviews code” to one that can do “almost everything developers and professionals do on a computer,” expanding the number of people who can build software and how they work. After testing the new model against a number of performance benchmarks, OpenAI claims it can “build highly capable and complex games and apps from scratch in days.”
OpenAI says the GPT-5.3 Codex is 25 percent faster than its previous model (GPT-5.2), and that it is the first model the company “helped create” and that its staff used an early version of the program to debug itself and evaluate its performance.
Note that the release of this new model comes on the heels of a new agent coding model released by our competitor Anthropic. In fact, OpenAI and Anthropic originally planned to release the two agent coding tools at the exact same time: 10 a.m. Pacific Time. However, shortly before the original release time, Anthropic moved its release date forward by 15 minutes and narrowly edged out OpenAI in the race to publish its model.
