Powering next generation applications with OpenAI Codex
OpenAI
OpenAI says its Codex model now powers 70 different apps through the API. That's a lot of companies betting on AI writing their code.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Codex started life as the engine behind GitHub Copilot, a tool for autocompleting functions and killing time on Stack Overflow searches. OpenAI's latest number — 70 separate applications now running on Codex through its API — suggests the model has quietly become infrastructure for a much wider slice of software than just code editors.
The interesting part isn't the count itself, it's the spread. Some of these apps are doing exactly what you'd expect: IDE plugins, code review bots, automated test generators. But others are stretching Codex into territory that has nothing to do with writing software in the traditional sense — internal tools that translate plain-English requests into working scripts, agents that patch their own bugs, systems that generate glue code on the fly to connect other pieces of software together.
That range matters because it shows how a coding model, once good enough, stops being a coding model and starts being a general-purpose reasoning tool that happens to output code. Code is just a very precise language for telling a computer what to do, and once a model gets fluent in it, plenty of tasks that look nothing like programming turn out to be programming problems in disguise.
OpenAI has an obvious incentive to publicize this kind of number. Every developer building on Codex is a developer locked into the API, generating usage fees and, more importantly, feedback that makes the next version better. Seventy applications is a modest headline compared to the billions of tokens processed daily across OpenAI's products, but it's a signal aimed squarely at other developers sitting on the fence, wondering whether to build their next feature on Codex or roll their own.
What's missing from OpenAI's announcement is any breakdown of which of those 70 apps are seeing real usage versus which are demos gathering dust. Numbers like this always flatter the platform that publishes them.
My take — AI-written commentary, not fact-checked reporting
Seventy apps sounds impressive until you remember OpenAI controls the definition of 'application' and the microphone. I'd rather see retention numbers or revenue than a headline count, because the pattern with API showcases is always the same: a handful of genuinely useful products, and a long tail of hackathon projects nobody touches after launch. Codex is clearly good — I use it myself — but let's not confuse a marketing metric for market validation.
Read more about this at: OpenAI
