OpenAI’s Codex crossed 20 million weekly active users this week, while Claude Code’s four-week growth rate has fallen to 5.2%. The coding agent market — the first AI product category with verifiable enterprise value — has entered a new phase: the easiest users have been captured, and the fight has shifted to enterprise revenue and the next generation of models.
The Growth Curves Just Crossed
Codex head Tibo confirmed on X that Codex passed 20 million weekly active users this week. CNBC-reported OpenAI internal data puts the combined “AI coding and work agent” products at roughly 20 million weekly actives. The run-up has been steep: through July, OpenAI streamed out milestones — 6M, 7M, 8M, 9M, and 10M users — then roughly doubled the figure in under a month.
On the other side, tracking firm TickerTrends shows Claude Code’s trailing four-week growth at 5.2% as of August 10, versus 20.8% for Codex — nearly four times faster. Between February and June the curves were reversed: Claude Code was the rocket and Codex was chasing.
Where Claude Code Slowed Down
Claude Code’s annualized revenue run-rate tells the story. It went from roughly $2.5B in February to over $6B by end of March, past $10B by end of April, and above $14B by June — a fivefold jump in four months. Then July arrived. Weekly estimates barely moved: $14.26B on July 6, $14.47B on July 20, a slight dip to $14.35B on July 27, $14.5B on August 3, and $15.12B by August 10.
Revenue did not collapse; it stopped accelerating. The reason is structural. Claude Code captured the easiest converts first: developers sit at computers with clear tasks, test suites, and verifiable output — the ideal agent use case. Once a product is good, natural early adopters flood in and build habits. Anthropic’s own June research found Claude Code users average about 20 hours of weekly usage. At that penetration, sustained high-double-digit growth is no longer plausible.
Lock-in is real too. The Information’s late-July enterprise reporting found many engineers resist switching even as prices rise. Across 200+ companies tracked by Weave, per-user cost roughly doubled in the first half while active users tripled. At Workato, Claude Code accounts for 80% of AI coding usage versus Codex’s 20%.
Why Codex Is Accelerating
Codex’s acceleration comes from two moves. First, distribution: OpenAI folded Codex into the ChatGPT entry point — the desktop app now exposes Chat, Work, and Codex with a shared agent quota. Second, a broader audience: knowledge workers now make up about 20% of Codex users — analysts, marketers, operations, designers, researchers, investors — and that segment is growing more than three times faster than developers.
The numbers deserve a caveat. TickerTrends does not track Codex revenue directly; it estimates $8.83B by assuming 45% of OpenAI’s enterprise ARR belongs to Codex, so the 20.8% growth figure is a model estimate, not a disclosed metric. What is verified: OpenAI’s own user data (5M+ weekly actives on June 2, up 6x since the February desktop launch) and management repeatedly framing Codex as a revenue accelerator.
Expand Outward vs. Consolidate Inward
The two companies are now running opposite playbooks. Anthropic is expanding Claude Code outward: taking the verified workflow and pushing it into Cowork, Claude Tag (which Anthropic describes as “an evolution of Claude Code”), and broader knowledge work. OpenAI is consolidating inward: pulling Codex into ChatGPT, sharing one quota, and funnelling a consumer base into agent usage.
This is the key structural insight. Anthropic is betting that a proven execution layer can spread from developers to every knowledge worker. OpenAI is betting that a consumer funnel plus a unified entry point can convert massive user numbers into enterprise revenue. Both are testable hypotheses about where AI agents create durable economic value — the same question behind autonomous cloud agents and the broader security of AI coding assistants.
The Company-Level Reality: Anthropic Still Leads
Zoom out and the picture flips. At the end of 2025, OpenAI’s run-rate (~$20B) was roughly double Anthropic’s (~$9B). By end of July, Anthropic’s run-rate exceeded $65B versus OpenAI’s $40B+ — Anthropic is now ahead by about 60%. In Q2 alone, Anthropic posted $11.6B in revenue, its first quarterly win over OpenAI’s $6.7B, though accounting treatments differ.
Claude Code is not even Anthropic’s main revenue source. Model sales to enterprises and third-party products — Cursor, Cognition, GitHub Copilot — form the core, and Anthropic expects API plus related enterprise applications to remain over 80% of revenue through 2028. Claude Code’s role is strategic: it proves the model advantage in a flagship use case and reinforces demand for the underlying Claude models. This is the same revenue arithmetic behind Anthropic’s $2T IPO run-rate story.
The Real Battleground Is the Next Model
Product entry points, enterprise sales, and user habits all amplify an advantage, but the core asset remains the model. OpenAI has an unreleased Astra, reported to show a step-change in agentic coding and cybersecurity — to the point that OpenAI slowed development for additional safety testing. Anthropic is rumored to be internally testing Fable 5.1. If either next-generation model opens a clear capability gap, today’s product leadership could be rewritten, just as the scaling debate continues to reset assumptions about model design.
What to Watch Next
Three signals matter more than user counts. First, quarterly revenue conversion: can Codex’s user surge become high-value enterprise contracts, especially through Codex Labs’ push with large system integrators? Second, model releases: Astra and Fable 5.1 will reset the coding capability bar. Third, knowledge-work expansion: whether Anthropic’s outward play and OpenAI’s consumer funnel actually convert non-developers into paying agent users.
FAQ
How many users does Codex have now? Codex passed 20 million weekly active users this week, per OpenAI’s Codex head; OpenAI’s internal data puts its combined coding and work agent products at roughly 20 million weekly actives.
Is Claude Code slowing down? Its trailing four-week growth rate fell to 5.2% as of August 10, but annualized revenue still rose to about $15.12B. Growth slowed, not revenue itself.
Who is winning the coding agent war? At the company level Anthropic still leads — run-rate above $65B versus OpenAI’s $40B+ — but Codex is growing faster in users. The decisive factor will be the next model generation.
