On Monday, OpenAI launched Codex, an agent-based coding tool offered to software developers. OpenAI also released a new Codex supercharger model today: the GPT-5.3 Codex.
The company says this model transforms Codex from an agent that can only “write and review code” to one that can do “nearly everything that developers and computing professionals do, expanding the possibilities of who can build software and how work gets done.” After testing its new model against a series of performance benchmarks, OpenAI claims it can create “highly functional complex games and applications from scratch within days.”
OpenAI says the GPT-5.3 Codex is also 25 percent faster than its predecessor (GPT-5.2) and that it was the company’s first model that “was instrumental in building itself,” meaning that company employees used early versions of the program to debug and evaluate how it worked.
The release of the new model follows closely on the heels of the new agent coding model released by its competitor Anthropic. OpenAI and Anthropic originally planned to release their two agent coding tools at exactly the same time: 10:00 a.m. PST. However, not long before the original release time, Anthropic pushed the release date up 15 minutes, narrowly beating OpenAI in the model promotion race.