Codex hit 7M users while Claude Code went silent — here's what the usage gap tells us

OpenAI disclosed yesterday that Codex Desktop now has 7 million users, up from under 700K six months ago. The climb accelerated: they added roughly 1 million users in the past 24 hours alone.

Anthropic launched Claude Code in March. They haven't published a single usage number since.

That gap is the story.

The Codex trajectory

Codex shipped in November 2025 as a paid add-on to ChatGPT Pro. By January it was available standalone. The early adoption was slow — developer tools always are. Then two things happened: GPT-5.6 Sol shipped in April with meaningfully better code output, and OpenAI started bundling Codex Desktop into enterprise seats at no extra cost.

The 10× growth from 700K to 7M happened in six months. Most of it came in the last 90 days. The 1M-user-per-day velocity suggests they're now in the viral-spread phase where every team that adopts it brings in three more teams the following week.

Codex runs in a desktop app. It has filesystem access, can spawn terminals, and integrates with VS Code and JetBrains IDEs. It's not a web IDE. It's a persistent copilot that sits next to your editor and executes plans across multiple files. That architecture matters because it removes the context-switching cost that killed earlier attempts at AI coding tools.

Claude Code's silence

Claude Code launched four months ago. Anthropic promoted it heavily at the AIE World's Fair in March. The demos were good — Opus 4.5 was the best reasoning model at the time, and the agentic loop they built on top of it handled multi-file edits well.

But there's been no follow-up. No usage milestones, no case studies, no "we processed X million coding sessions" announcements. Anthropic typically shares metrics when the numbers look good — they announced 300M Kagi searches processed within weeks of that integration, and they were vocal about Claude passing ChatGPT in certain segments last year.

The silence suggests the adoption curve isn't where they need it to be.

There are a few possible explanations. Claude Code might be structurally harder to distribute — it's a web-first tool, and web-based coding environments have a narrower wedge than desktop apps that hook into existing workflows. Or Opus 4.8's regression on tool use (which Armin Ronacher documented two weeks ago) might have hurt the product's reliability at exactly the wrong moment. Or the enterprise sales motion is slower because Anthropic doesn't have the ChatGPT install base to bundle on top of.

All of those could be true. But the simplest read is that Codex is winning because it shipped first with a better model at the moment of mass-market readiness, and network effects in developer tools compound fast.

What this means for the coding-agent market

The 7M number is a floor, not a ceiling. Codex is still adding a million users every day or two. If that pace holds for another month, they'll be at 10M+ by mid-August. At that scale, OpenAI will have more live coding sessions per day than GitHub Copilot had at its peak in 2024.

That has two effects. First, it locks in Codex as the default coding agent for any new developer joining the workforce. The same way "Google it" became the verb, "Codex it" is becoming the reflex for "write this function." Second, it gives OpenAI an overwhelming training advantage — 10M users generating multi-turn coding sessions is a data asset no competitor can match.

Claude Code can still win in verticals. If Anthropic focuses on specific use cases (financial services code review, healthcare data pipelines, compliance-heavy environments where Codex's speed matters less than auditability), they can carve out defensible niches. But the broad horizontal market for "AI that writes code" is consolidating around Codex faster than anyone expected six months ago.

The question now is whether Anthropic doubles down or pulls back. Silence could mean they're retooling. Or it could mean they've already decided the fight isn't worth it and they're reallocating resources to markets where Claude has clearer differentiation.

Either way, the Codex numbers are the new baseline. Any coding agent that launches from here forward will be measured against 7 million users and rising.