On Monday, OpenAI launched Codex, an agentic coding instrument marketed to software program builders. In the present day, OpenAI additionally launched a brand new mannequin designed to turbo-charge Codex: GPT-5.3 Codex.
The corporate says that the mannequin transforms Codex from an agent that may merely “write and evaluate code” to at least one that may do “practically something builders and professionals do on a pc, increasing who can construct software program and the way work will get accomplished.” Having examined its new mannequin in opposition to a lot of efficiency benchmarks, OpenAI claims that it may well create “extremely useful complicated video games and apps from scratch over the course of days.”
OpenAI says that GPT-5.3 Codex can be 25 % quicker than its earlier mannequin (GPT-5.2) and that it was the corporate’s first mannequin that “was instrumental in creating itself,” which means that the corporate’s employees used early variations of this system to debug itself and consider the way it was performing.
The brand new mannequin launch notably follows sizzling on the heels of a new agentic coding model launched by its competitor, Anthropic. Certainly, OpenAI and Anthropic had initially deliberate to launch their two agentic coding instruments at the very same time: 10 a.m. PST. Nevertheless, not lengthy earlier than the unique launch time, Anthropic moved its launch date up by quarter-hour, barely besting OpenAI within the race to publicize the fashions.
