Codex hits 20M as Tibo announces a banked reset amid usage-limit probe

Codex has crossed 20 million active users, according to Tibo Sottiaux, OpenAI’s Codex product lead. In a post on August 21, Tibo said OpenAI would credit every Codex and ChatGPT Work user with a banked reset during the day, while an investigation into reports of usage limits draining faster than expected continues.

A 20M-user milestone and a saved reset

The announcement combines a growth milestone with a practical entitlement change. Tibo said the reset would be banked, meaning users can keep it for a later moment instead of having it applied immediately. He also teased “other good news” for later, without specifying what that would be.

The wording matters: this is an announcement of the planned crediting process, not a guarantee that every account will show the reset at the same time. Users should check the usage summary in their Codex or ChatGPT desktop experience and allow for staged rollout or account-specific eligibility.

What OpenAI’s usage guidance confirms

OpenAI’s current Codex help page says Codex and ChatGPT Work can draw from a shared allowance and credit pool where those features are available on a plan. It also explains that a full banked reset replenishes both the five-hour and weekly usage windows. Redeeming one moves the weekly reset date to roughly seven days after redemption, so using it is a scheduling decision rather than a disposable bonus.

The same page says one-time resets may be offered to eligible users and that eligibility, affected limits, redemption rules, and expiry can vary. OpenAI’s separate promotion terms likewise describe rate-limit resets as benefits limited to the eligible Codex or ChatGPT Work experience, rather than API credits or transferable balances. Those qualifications are important because Tibo’s broad announcement and account-level rollout details may not appear identically for every plan or workspace.

Why faster usage is still under investigation

Tibo said OpenAI was not currently seeing anything abnormal in its checks, but was taking reports of faster-draining limits seriously and continuing an investigation. He also described a specific usage pattern involving subscription-to-API sharing through third-party services as something OpenAI’s fraud-prevention systems flag; ordinary Sign in with ChatGPT use through official or supported open-source clients was described separately.

That leaves the central user question open: whether some reports reflect model behavior, task complexity, account or workspace conditions, unsupported traffic patterns, or a measurement issue. OpenAI’s help documentation already notes that usage varies with the model, task complexity, context, reasoning effort, speed, tools, and long-running work. Tibo’s post does not establish that subscription limits were broadly reduced, and this article does not treat the investigation as a confirmed platform-wide bug.

What Codex users should check next

If the banked reset is relevant to an upcoming project, check the account’s usage summary for the reset count, redemption controls, and any expiry language before consuming it. Record the displayed allowance, reset date, client, model, and approximate task pattern if usage still appears abnormal; OpenAI asks users contacting Support to include those details and remove sensitive information from screenshots.

For teams, the immediate operational takeaway is to treat the reset as temporary capacity, not as a change to the plan’s permanent limits. The 20M milestone shows how quickly agentic coding has expanded, while the unresolved usage investigation shows that capacity, accounting, model behavior, and abuse controls are now part of the same product experience.

Evidence and caveats

The 20M figure, the planned banked reset, and the ongoing investigation are attributed to Tibo Sottiaux’s official announcement. OpenAI’s current Help Center documentation independently confirms how banked resets and shared Codex/Work allowances work, while its promotion terms explain why availability and expiry can differ. The account-level result remains the authoritative source for whether a specific user has received the reset.

Relevant LinkLoot workflows for planning agent-heavy work are collected in the AI workflow automation guide.