Codex vs. Claude Code: Is OpenAI’s Rapid Growth a True Comeback or a Statistical Trick?

Imagine a high-stakes chess match where two AI programming agents have been competing for nearly a year. Anthropic’s Claude Code, playing black, dominated the first half of 2026 with early enterprise penetration and strong developer community reputation. OpenAI’s Codex, the white player, started later but is charging ahead at an astonishing pace, backed by ChatGPT’s massive user base. The burning question: Has Codex truly begun to overtake Claude Code? To answer this, we need to look beyond surface metrics and understand the real context behind the numbers.

Recent data has turned heads. OpenAI CFO Sarah Friar announced during an August 2026 internal all-hands meeting that Codex’s weekly active users (WAU) had crossed the 20 million mark. Third-party platform TickerTrends added more fuel: As of August 10, Claude Code’s annual recurring revenue (ARR) stood at $15.1 billion, while Codex’s ARR reached $8.83 billion. The growth gap is striking: From July to August, Claude Code’s ARR grew by less than $900 million (from $14.26B to $15.12B), whereas Codex’s ARR jumped by nearly $3 billion (from $5.88B to $8.83B), with a 20.8% user growth rate—four times Claude Code’s 5.2%.

But there’s a critical catch in these numbers. OpenAI’s 20 million WAU isn’t just for Codex; it combines users of Codex and ChatGPT Work, an integrated office product. This is like merging two separate products’ user counts to inflate the figure. Codex alone has over 5 million WAU (a sixfold increase since its desktop launch) but that’s far from the 20 million combined number. Unfortunately, there’s no way to get a direct, separate WAU comparison for Claude Code currently.

Claude Code still holds key advantages. Anthropic’s Q2 revenue hit $11.6 billion—surpassing OpenAI’s $6.7 billion—with Claude Code as a core growth driver. Around 75-85% of Anthropic’s revenue comes from enterprise API calls, where Claude Code is the primary entry point for developers. OpenAI, by contrast, gets over 65% of its revenue from consumer subscriptions, lagging in enterprise penetration. Claude Code also offers cost efficiency: Its higher task completion accuracy reduces repeat calls and manual checks, leading to lower overall costs for businesses. This explains why over 1,000 enterprise clients (including Amazon, Google, and JPMorgan) spend more than $1 million annually on Anthropic’s services. A Jefferies test in August found that in mixed office and programming tasks, Codex and Claude Code are on par—no clear winner in pure programming.

Codex has gained momentum via integrated access and aggressive pricing, pulling Claude Code from its earlier 80% monthly growth to single digits. But Claude Code remains strong in the profitable enterprise segment, with high customer loyalty and its first positive operating profit. However, Claude Code faces mounting pressure: Codex is grabbing users, Chinese open-source models like DeepSeek and Kimi are undercutting it in the mid-to-low market with much lower prices, and Harness’s open-source framework could disrupt the game anytime. Its growth heyday seems over, and its product barriers are eroding.

Sources

• Toutiao Article: https://m.toutiao.com/article/7676774213277073920/
• TickerTrends (ARR tracking data for AI programming agents)
• OpenAI Internal All-Hands Meeting (August 2026, comments by CFO Sarah Friar)
• Jefferies Global AI Agent Test (August 2026)