Recently, a bunch of people have been hyping up how powerful GPT 5.3 is, but everyone might be overlooking one thing: the 5.3 version is currently only available to Codex users. API and ChatGPT web users can’t access it—it’s just a small group of people using it. When a model is only open to a limited number of users, compute resources are abundant, inference runs more thoroughly, and latency is lower. Once it’s fully opened up to millions of API and web users, to control costs and latency, there are usually some trade-offs, like more aggressive batching, tweaking sampling parameters, or even replacing it with a distilled version. This is the context behind those “GPT 5.3 is amazing” experience posts. Opus 4.6 is currently open for API, web, and Claude Code users. People all over the world are using it simultaneously, and it’s handling real, large-scale traffic. Is it fair to compare the experience of a model serving only a small number of users to one that’s fully open? Sam Altman is smart—he locked the most powerful model in Codex first, creating a wave of “scarcity” and good word-of-mouth. Coding tools are super profitable, and Codex is a competitor to Claude Code, so he has to create the best user experience. So those posts hyping up 5.3 are, to some extent, just a temporary bonus. Opus 4.6 is competing with you on the real battlefield of full-scale deployment. When GPT 5.3 enters the same battlefield, then we can talk about who’s stronger.