What does OpenAI's latest report reveal about Codex's user growth, its expansion beyond developers into knowledge workers, the product updates that broadened its capabilities, and the competitive context against Anthropic's Claude franchise?

Expansion into knowledge workers

• Knowledge workers now account for ~20% of weekly users and are adopting Codex more than 3× as fast as developers.
• Their primary uses: creating reports, spreadsheets, presentations, contracts, research documents, and other non-code work products.
• OpenAI says the data signals "a broader shift is underway" — Codex is transitioning from a developer tool into a general enterprise productivity platform.

Key product updates that broadened capabilities

• Desktop app launch (Feb 2026): A full desktop application that triggered the 6×+ user surge.
• GPT-5.4 / GPT-5.5 model upgrades (March–May 2026): New default models brought substantial benchmark improvements.
• Computer Use (Windows/Mac): Codex can now see, click, and type in desktop applications, enabling it to test, debug, and automate GUI-based workflows.
• Plugins platform (June 2, 2026): Enterprise plugins for banking, sales, design, analytics — directly targeting white-collar knowledge work.
• Goal Mode (GA): Autonomous, long-running task execution without continuous user supervision.
• Sub-agents & parallel execution: Codex can spawn sub-agents for multi-tasking, a structural advantage over Claude Code.
• Appshots, @mentions, MCP shortcuts, remote control, memory, image generation: All added throughout 2026, making Codex a "tool for (almost) everything".

Competitive context vs. Anthropic's Claude franchise

• Claude Code (Anthropic's competing agent) launched in February 2025, three months before OpenAI's Codex relaunch in May 2025.
• Benchmark parity: By May 2026, the two products are roughly at parity on leading benchmarks — GPT-5.5 narrowly leads SWE-bench Verified (88.7% vs 87.6%) and Terminal-Bench 2.0 (82.7%), while Claude Opus 4.7 leads SWE-bench Pro (64.3% vs 58.6%).
• Philosophical differences: Claude Code runs locally with a 1M-token context window (4× larger than Codex's 272K) and is preferred for complex multi-file refactoring, but burns 3–4× more tokens per task. Codex runs in a cloud sandbox, uses significantly fewer tokens, and supports native parallel agent execution.
• Enterprise expansion race: OpenAI launched Codex plugins for banking/sales/analytics on June 2, 2026, directly responding to Anthropic's earlier moves into white-collar automation — "After Anthropic, OpenAI launches Codex plugins to automate banking, sales and other white-collar jobs".
• Both are actively targeting the same enterprise agent market, with Codex leveraging its broader knowledge-work pivot and Claude Code leaning on local execution, security, and longer-context reasoning.