GPT‑5.5 rollout doubled Codex API sales, OpenAI says

OpenAI’s Codex story just changed shape. This is no longer only about autocomplete for engineers or a nicer code review tool. After the GPT‑5.5 rollout on April 23, OpenAI says Codex API revenue doubled within a week, and the company is now talking about Codex as something that can handle coding, browser work, and broader knowledge tasks across a company. ### Why is GPT‑5.5 the hinge here? GPT‑5.5 is the model OpenAI built for longer, messier work — coding, research, spreadsheets, software operation, and multi-step tasks that need planning and tool use instead of constant hand-holding. In the launch post, OpenAI also said GPT‑5.5 uses fewer tokens on the same Codex tasks while matching GPT‑5.4 latency, which helps explain why a model upgrade could immediately move revenue as well as usage. ### What changed inside Codex itself? Codex got broader at almost the same moment. OpenAI’s changelog says the app now includes browser use for local development pages, automatic approval reviews for some actions, and a push toward “computer use” so Codex can click, type, and verify work in apps instead of only editing code. That makes the product feel less like an IDE sidebar and more like an agent that can move across tools. ### Why does “API revenue doubled” matter? Because it hints that customers are paying for automation, not just experimenting. The public claim came from Alexander Embiricos, OpenAI’s Codex product lead, who said the latest Codex model’s API revenue doubled within a week and was growing 2x faster than any prior OpenAI release. That is a very different signal from consumer buzz — it suggests teams found workflows worth wiring into production fast. ### Is this still mostly a developer tool? Not really — or at least that is not how OpenAI is framing it now. In the same interview, Embiricos described the path as easy tasks first, then harder ones, then full automation, and he talked about Codex being useful for sales, marketing, finance, and data work, not just engineering. OpenAI’s own GPT‑5.5 launch post makes the same move by grouping agentic coding with computer use and knowledge work. ### What does enterprise adoption actually look like? OpenAI says Codex crossed 3 million weekly developers in early April and then topped 4 million weekly users by April 21. It also launched Codex Labs and named Accenture, Capgemini, CGI, Cognizant, Infosys, PwC, and procurement, compliance, and change management. ### So what are customers buying now? Basically, a software worker that can read a repo, make edits, run commands, inspect a rendered page, and sometimes operate a desktop interface. OpenAI’s docs pitch Codex for implementation, refactors, debugging, testing, validation, and knowledge-work artifacts, while the product page leans into “weeks of work in days” and parallel agents important. ### What’s the catch? The catch is governance. The more Codex can act across repositories, browsers, and apps, the less the buying conversation is about model quality alone. Enterprises now have to care about approval flows, internet access settings, environment controls, auditability, and whether an agent can be trusted around sensitive systems. OpenAI’s own docs already surface those controls, which tells you where adoption friction is moving. ### Bottom line? The interesting part is not just that GPT‑5.5 made Codex better. It is that the upgrade seems to have pushed customers from “help me code faster” toward “take this workflow and run it.” If that holds, Codex is becoming an enterprise automation product that happens to start in software development — and that is a much bigger market.