OpenAI Codex has 4 million weekly active users and is generating just over $1 billion in annualised recurring revenue — the first AI coding agent to achieve that revenue milestone and one of the fastest product-to-billion-dollar trajectories in enterprise software history. These figures, disclosed by OpenAI in the context of the Codex product expansion announced in mid-May 2026, place Codex alongside Claude Code as the two leading agentic coding agents in the market. Claude Code reached approximately $2.5 billion in ARR by February 2026, doubling from $1 billion at end-2025 — so Codex is behind in ARR but ahead in weekly active user count among the publicly disclosed figures. The competitive dynamics between Codex and Claude Code are the most closely watched product-level AI competition in enterprise software, and the week that OpenAI filed its confidential IPO prospectus is also the week that Codex’s scale became fully visible to the market.
What Codex Can Do — The Full Autonomous Capability Stack
Codex in May 2026 is not the GitHub Copilot autocomplete that many developers still associate with the OpenAI coding product brand. It is a fully agentic coding system capable of autonomous multi-step software engineering tasks — reading a codebase, understanding its architecture, planning a change, writing the code, running tests, iterating on failures, and submitting a pull request — without requiring human input at each step. OpenAI describes Codex as a command centre for managing multiple coding agents, parallel threads, worktrees, skills, automations, and long-running software tasks simultaneously. The system is available in the ChatGPT app, through the OpenAI API, as an IDE extension for popular development environments, and through a dedicated CLI.
The production deployment list that OpenAI has disclosed demonstrates the breadth of enterprise adoption: Harvey (legal AI), Sierra (customer service AI), Ramp (fintech), Cisco Meraki (networking), Duolingo (education technology), and dozens of other organisations have integrated Codex into their production software engineering workflows. These are not pilot programmes — they are live deployments where Codex is executing multi-step coding tasks in production environments with approval gates that allow engineering teams to review Codex’s proposed changes before they are merged. In our hands-on review of Codex’s published capability documentation and the production deployment evidence, the system’s most remarkable property is not any single capability but the integration depth: once a team has set up Codex with its codebase context, coding conventions, test infrastructure, and approval workflow, the switching cost to a competitor system is substantial.
“Codex is a command centre for managing multiple coding agents, parallel threads, worktrees, skills, automations, and long-running software tasks simultaneously. It’s not a coding assistant — it’s a coding agent that works while you sleep.” — OpenAI Codex official product documentation, May 2026
OpenAI Codex vs Claude Code — Head-to-Head Comparison (May 2026)
| Metric | OpenAI Codex | Anthropic Claude Code | Advantage |
| Weekly active users | 4 million | Not disclosed — est. higher (enterprise-heavy) | Codex — more disclosed WAU |
| Annualised revenue | $1 billion+ | $2.5 billion (February 2026 figure) | Claude Code — 2.5x larger ARR |
| SWE-bench Verified score | 88.7% (GPT-5.5) | 64.3% (SWE-bench Pro — different benchmark) | Difficult to compare directly — different test sets |
| Terminal integration | Yes — CLI available | Yes — terminal-native (primary interface) | Claude Code — deeper terminal nativism |
| Windows support | Yes — full sandbox architecture published May 13 | Yes — available | Codex — more detailed security documentation |
| Mobile management | Yes — Codex Mobile (iOS and Android) launched May 15 | Limited — web-focused | Codex — broader device coverage |
| IDE extensions | VSCode, JetBrains, and others | VSCode, JetBrains, and others | Tied |
| Remote SSH | GA — generally available | Available | Codex — confirmed GA status |
| Enterprise security docs | Full sandbox architecture published May 13 | Not publicly detailed at same level | Codex — more transparent security documentation |
Codex Mobile — Managing Agents From Anywhere
OpenAI launched Codex for ChatGPT mobile on iOS and Android on May 15, 2026, extending the agent management surface from desktop and terminal environments to smartphone. The mobile implementation is not a full coding environment — it is a management interface for Codex sessions already running on a laptop, Mac mini, development server, or remote environment. Developers can start work, inspect active threads, review outputs, approve commands, and follow terminal output, screenshots, diffs, and test results in real time from their phone. Files, credentials, and compute remain on the machine where Codex is executing; the phone provides a control surface for directing and monitoring long-running agent tasks that continue while the developer is away from their workstation.
The practical significance of Codex Mobile is not primarily about new capability — it is about coverage. An agentic coding system that runs 24-hour autonomous tasks on a server needs a management interface that is available when the developer is not at a desk. The ability to approve a Codex pull request from a phone, check whether a long overnight test suite passed, or redirect a running agent that has taken an unexpected path addresses the specific friction that autonomous coding agents create: they run continuously, but developers do not. Remote SSH, now generally available in Codex, extends this further: Codex can connect into approved remote development environments with company dependencies and security policies, allowing agents to operate within existing enterprise infrastructure rather than requiring a dedicated Codex cloud environment.
“If Codex gets stuck on something, it surfaces what it’s unsure about. If it crosses a boundary you’ve set, it comes back to you for approval. The human is always in the loop on the decisions that matter.” — OpenAI Codex product documentation on the auto-review approval layer, May 2026
The $1 Billion ARR Milestone — What It Means for the IPO
Codex passing $1 billion in annualised revenue is the most important Codex data point for OpenAI’s IPO prospectus. The S-1 that OpenAI files publicly in late July 2026 will need to tell a coherent story about product-market fit in the AI era — and a $1 billion ARR coding agent with 4 million weekly users is a cleaner, more defensible product story than a general-purpose chatbot whose revenue is driven by a mix of consumer subscriptions, enterprise contracts, and API usage that is harder to attribute to specific product value. Codex’s revenue is primarily API-based — enterprises pay per-token or per-completion for Codex API calls, meaning the revenue is directly tied to productive use rather than to subscription retention that may or may not correlate with actual usage.
The competitive threat to Codex comes primarily from Claude Code. Anthropic’s revenue trajectory — from $1 billion Claude Code ARR at end-2025 to $2.5 billion in February 2026 — is growing faster than Codex’s currently disclosed trajectory. The $1 billion figure for Codex is estimated to reflect end-2025 or early 2026 data, while Claude Code’s $2.5 billion figure is February 2026. If Codex is growing at a similar rate, its May 2026 ARR may be higher than $1 billion — but OpenAI has not disclosed the updated figure. The 4 million weekly active user count is the most current disclosed metric and is likely more up to date than the revenue figure. For the IPO prospectus, both Codex’s revenue trajectory and its user base are assets — but they are assets that will be directly compared to Claude Code’s disclosed metrics by every analyst who reads both companies’ public filings in succession.
Key Takeaways
• OpenAI Codex has reached 4 million weekly active users and over $1 billion in annualised recurring revenue — the first AI coding agent to achieve billion-dollar ARR, achieved within approximately 12 months of its May 2025 launch as a generally available agentic product.
• Codex is deployed in production engineering workflows at Harvey, Sierra, Ramp, Cisco Meraki, Duolingo, and dozens of other organisations — agentic multi-step coding with approval gates, not autocomplete assistance.
• Codex Mobile launched May 15, 2026 on iOS and Android, providing a management interface for long-running Codex agent sessions on remote hardware — allowing developers to approve commands, review outputs, and redirect running agents from their phones.
• Remote SSH is now generally available in Codex, enabling agents to connect into existing enterprise development environments with company dependencies and security policies rather than requiring a separate cloud sandbox.
• OpenAI’s full Windows sandbox architecture for Codex — published May 13, 2026 — establishes a detailed security documentation standard for enterprise AI coding agent deployment that competitors including Claude Code have not yet matched at the same transparency level.
• The $1 billion Codex ARR milestone is a critical component of OpenAI’s IPO narrative filed May 22 — providing a defensible product-specific revenue story alongside the broader ChatGPT consumer and API business in a prospectus that must justify an $852 billion valuation.
Conclusion
OpenAI Codex’s 4 million weekly users and $1 billion ARR in approximately 12 months of general availability is one of the fastest product-to-market milestones in enterprise software history. The product timeline — from autocomplete assistant to autonomous multi-step agent with Windows sandbox security, mobile management, remote SSH, and approval-gated enterprise deployment — is the clearest available evidence that the shift from code assistance to code agency is real, commercial, and growing. The competitive context with Claude Code remains the most important variable: Anthropic’s agent is generating 2.5x the ARR at $2.5 billion, which means Codex must either demonstrate a faster growth trajectory, a superior enterprise integration story, or a differentiated product capability to avoid the risk of being positioned as the challenger to Claude Code rather than the market leader. The Codex Mobile launch, the Windows security documentation, and the 4 million weekly user milestone are all components of the same argument: Codex is a complete enterprise-grade agentic coding platform, not a prototype. The IPO will determine how the market prices that argument relative to the platform that is currently generating more revenue.
Frequently Asked Questions
How many users does OpenAI Codex have?
OpenAI Codex has 4 million weekly active users as of May 2026. This figure was disclosed alongside the Codex product expansion announcements in mid-May 2026 and represents users who interact with Codex through ChatGPT, the OpenAI API, IDE extensions, and the CLI each week.
How much revenue does OpenAI Codex generate?
OpenAI Codex generates just over $1 billion in annualised recurring revenue as of the most recent disclosed figure. This makes it the first AI coding agent to reach billion-dollar ARR. For comparison, Claude Code reached $2.5 billion in ARR by February 2026 — Codex trails on revenue but leads on disclosed weekly active users.
What can OpenAI Codex do autonomously?
Codex can autonomously read a codebase, understand its architecture, plan changes, write code, run tests, iterate on failures, and submit pull requests without step-by-step human instruction. It manages multiple parallel coding threads simultaneously and uses an approval-gate system where high-risk actions (boundary-crossing commands, network access) require explicit human confirmation before execution.
What is Codex Mobile?
Codex Mobile is an iOS and Android app launched May 15, 2026 that provides a management interface for Codex sessions running on a laptop, server, or remote development environment. Developers can approve commands, review outputs, check thread progress, and redirect running agents from their phones. Files and compute remain on the machine where Codex is executing; the phone is a control surface, not an execution environment.
How does Codex compare to Claude Code?
Codex has more disclosed weekly active users (4 million) but lower disclosed ARR ($1 billion versus Claude Code’s $2.5 billion in February 2026). Codex is GPT-5.5-powered with 88.7% SWE-bench Verified score; Claude Code is Claude-powered with a 64.3% SWE-bench Pro score (different benchmark). Both are enterprise-grade agentic systems with IDE integration, terminal access, and approval-gate security. Codex has published more detailed Windows sandbox security documentation.
References
OpenAI. (2026, May). Codex product page and documentation. https://openai.com/codex
OpenAI. (2026, May 13). Building a safe, effective sandbox to enable Codex on Windows. https://openai.com/index/building-codex-windows-sandbox/
TestingCatalog. (2026, May 16). OpenAI brings Codex to ChatGPT mobile app for iOS and Android. https://www.testingcatalog.com/openai-brings-codex-to-chatgpt-mobile-app-for-ios-and-android/
eWeek. (2026, May 16). OpenAI launches Codex in ChatGPT mobile for its 4M weekly users. https://www.eweek.com/news/openai-codex-mobile-chatgpt-app/
Build Fast with AI. (2026, May 21). AI news today — May 22, 2026: 12 biggest stories. https://www.buildfastwithai.com/blogs/ai-news-today-may-22-2026
TechTimes. (2026, May 16). OpenAI unifies ChatGPT, Codex, and developer API under Brockman. https://www.techtimes.com/articles/316730/20260516/openai-unifies-chatgpt-codex-developer-api-under-co-founder-brockman.htm
OpenAI Developers. (2026). Codex concepts and sandboxing documentation. https://developers.openai.com/codex/concepts/sandboxing
