OpenAI Codex Reaches 20 Million Users and Gives Paid Users a Banked Reset

GPT-5.6 launched on July 9, during the period when Codex and ChatGPT Work were reporting rapid milestone gains. The evidence available here does not establish that the launch alone caused the subsequent increase. Product availability, the launch of ChatGPT Work, and broader adoption may all be part of the story, but the causal breakdown has not been demonstrated.

What a banked reset actually does

A banked reset is best understood as an on-demand recovery of depleted usage. Instead of waiting for the normal five-hour or weekly allowance to replenish, an eligible user can save the reset and activate it when a project requires more capacity.

That makes the benefit especially useful for users who reach their limit early in a work cycle. It addresses the immediate “I’m blocked” problem without forcing users to wait or change plans.

The reset does not appear to mean:

• unlimited Codex or ChatGPT Work usage;
• a permanent increase to a plan’s weekly allowance;
• a refund of API credits; or
• a change to the underlying rate-limit schedule.

Those distinctions matter because OpenAI’s documentation says the promotional GPT-5.6 Sol pricing applies to eligible purchased-credit usage, while included plan usage and five-hour and weekly limits remain unchanged.

Why the giveaway matters amid quota complaints

The banked reset offers practical relief as users report that heavy Codex sessions can exhaust their allowances before the end of a normal week. Complaints have focused on the visibility of usage meters and on whether compute-intensive workflows—including background or helper features—consume more allowance than users expect.

The available evidence does not establish that auto-review or helper subagents are definitively responsible for widespread quota depletion. Nor does it confirm the existence of a reported internal “warroom.” Those claims should remain separate from the confirmed milestone and reset announcement.

In that context, the giveaway is a customer-experience intervention rather than a redesign of Codex’s usage model. It gives paid users a safety valve while leaving the metering, quota, and cost-control systems in place.

The separate $8 pay-to-reset test

A separate report described an $8 “pay to reset” option appearing for at least one ChatGPT Plus user. OpenAI had not formally announced that feature in the evidence available here, so it should be treated as a reported test rather than a generally available product policy.

That reported test and the free banked reset point in the same direction: users may be willing to pay for continuity when an AI coding workflow is interrupted. But they are not the same benefit. Sottiaux’s August announcement described a banked reset granted to users, while the $8 option was reported as a possible paid mechanism for obtaining an immediate reset.

GPT-5.6 Sol is cheaper for credits—but subscriptions are unchanged

OpenAI also announced a temporary reduction of more than 20% in GPT-5.6 Sol API and eligible credit pricing for three months. The promotion applies to the API and rolls out to eligible ChatGPT Work and Codex credit usage; it does not reduce flat-rate Pro, Plus, or Business subscription prices or change included subscription limits.

The distinction is important for users deciding whether the reset solves their cost problem. Lower credit prices can make additional usage more affordable for credit-based workflows, but subscription users still face the same included allowances and reset windows documented in the Codex rate card.

The broader product trade-off

Codex’s August milestone shows how quickly an AI coding product can scale when it becomes part of a broader work platform. It also exposes the tension between adoption and compute economics: more users and more agentic workflows increase demand, while fixed quotas and metered credits help control capacity and cost.

The banked reset is a useful short-term compromise. It rewards adoption and reduces the frustration of being stopped at an inconvenient moment, but it does not answer the longer-term questions around quota transparency, background-task accounting, or how much capacity paid users should expect from their plans.

For now, the clearest takeaway is simple: Codex has reached a reported 20 million-user milestone, and OpenAI is offering paid users a one-time way to recover depleted usage—without changing the recurring limits that govern the service.