Sam Altman says OpenAI is resetting Codex's usage limits “to celebrate 3M weekly Codex users” and will reset them for every 1M new users until it reaches 10M
To celebrate 3 million weekly codex users, we are resetting usage limits. We will do this every million users up to 10 million. Happy building!
Context & Ripple Effects
Codex had already been broadened beyond its initial Pro-only availability through its expansion to Plus users and optional internet access. The limit reset turns user-growth milestones into a distribution lever rather than treating access solely as a fixed subscription entitlement.
Related coverage later places the move in a fast-adoption arc: Codex surpassed 5 million weekly active users and OpenAI reported sharply wider internal use, including among non-developers. At the same time, paid ChatGPT tiers were differentiated by substantially higher Codex allowances, making usage limits a core product-control mechanism rather than a minor setting.
First-order effects
- Current Codex users receive more room to use the service immediately, while OpenAI exchanges foregone near-term rationing for greater product engagement at a stated adoption milestone.
- OpenAI makes capacity allocation visibly contingent on growth: further resets are tied to each additional million users on the path to 10 million.
Second-order effects
- The move increases pressure on coding-agent rivals to compete on usable task volume and access predictability, not only model capability.
- It also sharpens the contrast with OpenAI’s paid tiers, whose higher Codex usage allowances remain a monetization and demand-management tool even as milestone resets widen access.
Third-order effects
- If sustained, milestone-based limit changes would make frontier-agent access a more dynamic commercial policy, with providers continuously balancing adoption incentives against finite inference capacity.
- The broader market may segment around managed access tiers: generous baseline usage to build habit, then paid or controlled capacity for heavier workloads as agents spread beyond software developers.
The trend: This is one instance of coding-agent platforms using access and capacity policy—alongside model features—to accelerate managed-model adoption while preserving levers for monetization and compute control.