Did Codex Really Overtake Claude Code? The 6M (Now 7M) User Number, Fact-Checked

On July 12, 2026, OpenAI's Thibault "Tibo" Sottiaux — Head of Core Products — posted on X that Codex had "hit 6M active users." A day later, he updated it to 7M. Within hours, headlines like Codex surges to 6 million active users, overtaking Claude Code's 2 million were circulating, and How-To Geek ran with "Codex has dethroned Claude as the king of AI programming, and it's not even close."

There's a real story in there about Codex's growth — it's fast, and it's genuine. But the specific "overtook Claude Code" framing doesn't survive contact with Tibo's own wording, let alone independent data. Here's what the number actually says, and what search-interest data shows instead.

Where the number actually came from

Both of Tibo's posts are still up, and both say the same thing in the same breath: this is a combined figure for two different products.

"Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates: [...] We hit 6M active users, and are landing a usage reset in the next hour." — @thsottiaux, July 12, 2026

"Thank you to the 7M active users who are now using Codex and ChatGPT Work. We have added a banked reset to everyone's account to celebrate the milestone." — @thsottiaux, July 13, 2026

Codex is OpenAI's coding agent — the direct Claude Code comparison. ChatGPT Work is a separate, broader agentic product for general knowledge work (drafting, research, spreadsheets, browsing) that OpenAI has been folding into the same ChatGPT desktop app, positioned as its answer to Anthropic's Claude Cowork. They are not the same tool, and OpenAI has not published a Codex-only active-user number since the two were merged into one metric.

Note the gap: the Codex-only number stops being reported after March. Every figure from June onward describes a combined product surface, but coverage keeps citing it as if it measures the same thing as the March number — or as if it's directly comparable to Claude Code's separately-reported figures.

The comparison problem: two products vs. one, and a stale baseline

The "Claude Code's 2M" half of the headline has the opposite problem: it's an old, undated number doing service as a live baseline. Anthropic hasn't published a single, clean "Claude Code weekly active users" figure this week to compare against — but its most recent concurrent disclosures paint a different growth picture than a flat "2M":

Both companies are growing. But stacking OpenAI's freshest combined-product number against Anthropic's staleest single-product number and calling it "overtaking" is comparing a current sum of two things to an old count of one thing. It would look just as skewed run the other way — Claude Code's revenue growth against a six-month-old Codex snapshot.

What independent search data shows

Marketing claims are one thing; where developers actually spend their attention is another. Google Trends measures search interest, not product usage, but it's a genuinely independent, third-party signal neither company controls — and it tells a different story than the headlines.

Across the full three-month window, "Claude Code" search interest never once dropped below "OpenAI Codex" — it ran 2-5x ahead the entire time, including the exact week the "overtaking" story broke. "ChatGPT Work" barely registers as a search term at all, which is itself informative: whatever 7M people are doing inside that combined metric, they mostly aren't searching for the ChatGPT Work brand name specifically.

Two more data points are directionally useful but need a heavier caveat, because they measure whole products, not Codex or Claude Code specifically: on Similarweb, chatgpt.com pulls roughly 5.4B monthly visits against claude.ai's 947M (June 2026 snapshot) — reflecting ChatGPT's much larger general-consumer base, not a coding-agent-specific split. Same story on mobile and Chrome: the ChatGPT app has 8.6M App Store ratings to Claude's 198K, and the official "Claude in Chrome" extension has roughly 11M users against the largest Codex-branded Chrome extension we found, at under 10K. None of this isolates Codex or Claude Code — Codex now ships inside the general ChatGPT app, and Claude Code is a terminal tool with no App Store or Chrome Web Store footprint of its own — so treat these as background context on overall brand scale, not evidence about the coding-agent race specifically.

What's genuinely true about Codex's growth

None of this means Codex's growth is fake. Going from a reported sub-1M weekly active user base in February to 5M by early June to 6-7M by mid-July, per OpenAI's own numbers, is a real and fast curve — even accounting for the metric merging with ChatGPT Work along the way. Some of that is organic pull from a genuinely improved coding model; some of it is straightforward distribution: Codex now ships bundled into every ChatGPT Plus, Pro, and Business seat, so a meaningful share of "active user" growth is existing ChatGPT subscribers being auto-exposed to a feature that's already inside their app, not net-new developers switching allegiance from a competitor.

That distinction — organic adoption vs. bundled exposure — is exactly what a single combined "active users" number can't tell you, and exactly why it's a weak instrument for a head-to-head claim against a competitor that ships as a separate, opt-in CLI tool.

The more accurate headline

"Codex overtook Claude Code" implies one developer tool beat another on its own terms. What the data actually supports is narrower and, honestly, more interesting: OpenAI is winning the distribution race by bundling Codex into a product with a much larger existing installed base, while Anthropic is still winning the developer-attention race that Google Trends measures independently of either company's press releases. Those are two different competitions, and a company's own milestone tweet — however genuine the underlying growth — isn't a substitute for measuring both of them side by side.

Sources

Methodology

The search-interest, traffic, app-store, and browser-extension figures in this piece were pulled directly via Crawlora's own data APIs rather than taken secondhand from other write-ups — Google Trends, Similarweb, App Store, and Chrome Web Store endpoints, plus the X profile/post endpoints used to pull Tibo's posts verbatim. If you want to run your own version of this comparison, or track it as the numbers keep moving, test the same endpoints in the Playground — the how to scrape Google Trends and best Twitter/X scraper API guides walk through the exact calls, and pricing covers the free tier if you want to reproduce this yourself.

For the AI-coding-agent landscape more broadly, see web scraping with AI and how to choose a web scraping API.