OpenAI's Codex and ChatGPT Work agents reach 10M users

The headline claim, reported by Bloomberg, is that OpenAI's Codex and ChatGPT Work now have roughly 10 million people using them, roughly double where the pair sat when ChatGPT Work debuted on July 9. That is fast even by 2026 chatbot standards, which is why the number is worth pulling apart rather than just reposting.

The mechanics behind the growth are clearer than the headline. On July 9 OpenAI folded its standalone Codex coding agent into the ChatGPT desktop app and simultaneously launched ChatGPT Work, an agentic mode that hands finished documents, spreadsheets and web apps back to the user instead of chat responses. That release was timed with the public rollout of GPT-5.6. Codex on its own went from just over 1 million weekly users on February 5 to 7 million by July 13, and days later Sam Altman said combined usage of the two agentic products had risen 2.5x in a single week. The fastest-growing cohort, per OpenAI, is people who have never written a line of production code — the Work half of the pitch, not the Codex half.

Why this matters if you are choosing agent tooling: the frame has shifted. Six months ago Codex was a developer product competing with Cursor and Claude Code, and enterprise agent buying sat on a separate procurement track. Bundling Codex with ChatGPT Work into one desktop surface, with paid tiers unlocking hours-long autonomous runs, prices the two conversations together. Anyone comparing agent vendors now has to reason about a Codex-plus-ChatGPT-Work bundle rather than a coding assistant on its own.

The honest caveat is that the number is OpenAI's number, relayed through reporting. Take the 10 million, the 2.5x and the nearly-doubling framing as reported, not audited. The reporting doesn't split how many of those users are Codex versus ChatGPT Work, doesn't specify whether the count is weekly-active or cumulative, and doesn't break out paid versus free. Altman has also been resetting weekly usage caps at each million-user milestone, which tends to pull lapsed accounts back into the active bucket right when the flag goes up.

What is worth watching is which of the incumbents responds first. Anthropic's Claude Cowork, Microsoft's Copilot and the Cursor-style IDE plays now have a public benchmark to position against, and OpenAI has a large non-engineering cohort to defend.