OpenAI Codex’s reported 90M-install week was an npm spike, not 90M users

npm download counts are a high-noise proxy. A separate npm-focused analysis said the official npm page showed 195,488,301 weekly downloads for @openai/codex and warned that npm downloads can include automated installs, CI, short tests, re-fetches and distribution effects ^[34]. OpenAI’s own GitHub materials include a GitHub Actions workflow that installs Codex before running an automated changelog task, showing how repeatable automation can create package downloads without mapping one-to-one to new human users ^[40].

What likely drove the spike

1. GPT-5.5 gave developers a reason to retest Codex

OpenAI said GPT-5.5 and GPT-5.5 Pro became available in the API in an April 24, 2026 update ^[17]. The same announcement said GPT-5.5 excels at writing and debugging code, can carry more of the work itself and uses significantly fewer tokens to complete the same Codex tasks ^[17].

That makes GPT-5.5 a credible catalyst for renewed Codex experimentation. Developers had a fresh reason to reinstall, update, benchmark or wire Codex into local and automated workflows. The available sources do not quantify how much of the reported 90M spike came from GPT-5.5, so it should be treated as a major likely contributor rather than the sole proven cause.

2. Codex CLI and the SDK are built for easy distribution

Codex is not just a web assistant. OpenAI describes it as a coding partner for tasks such as pull requests, feature work, complex refactors and migrations, powered by its frontier coding models ^[44]. The CLI lowers the adoption barrier further because developers can install it globally with a one-line npm command and run it from the terminal ^[37].

OpenAI’s general-availability post also emphasized the Codex SDK, which lets teams embed the same agent that powers Codex CLI into their own workflows, tools and apps ^[18]. That is strong product distribution, but it also means installs can be triggered by scripts, CI jobs and repeated setup processes rather than one-time manual downloads ^[34][40].

3. Codex already had momentum before the headline

The npm surge did not appear in isolation. Fortune reported in March 2026 that, since early February, more than 1 million people had downloaded the Codex desktop app and Codex had more than 1.6 million active weekly users, according to OpenAI ^[7]. Moneycontrol separately reported that the standalone Codex app crossed 1 million downloads in its first week on Mac and that usage rose 60% week over week after the February 2 launch and GPT-5.3-Codex rollout ^[8].

OpenAI’s February 5, 2026 GPT-5.3-Codex announcement described that model as its most capable agentic coding model to date and said it was 25% faster than the previous version ^[22]. Those earlier milestones point to real adoption momentum before the later package-download spike, even though they are separate from the 90M npm-style metric.

4. AI coding-agent competition made npm a scoreboard

The Codex surge also fits the broader race among coding agents. An npm API snapshot collected on May 6, 2026 reported about 210.5 million last-month downloads for @openai/codex versus about 46.5 million for @anthropic-ai/claude-code, putting Codex roughly 4.5x ahead on that package-download window ^[33].

That comparison is useful as a developer-attention signal, but it has the same caveat as the 90M headline. npm downloads can show which tools are being pulled into environments at scale; they do not prove retention, productivity gains or the number of active human users ^[33][34].

How to read the metric without getting fooled

Bottom line

OpenAI Codex’s reported 90M-install week is best understood as a distribution shock: GPT-5.5, GPT-5.3-Codex, the Codex SDK, one-line CLI installation and competitive attention all gave developers and automated systems reasons to pull the package ^{[17][18][22][33][37]}.

The caveat is just as important as the spike. npm-style download activity is not the same as active usage. The safer conclusion is that Codex became highly visible in developer package infrastructure, not that 90 million developers actively used it in a week ^[1][34].