OpenAI is betting that packaging flexibility, not a headline price cut, will slow Anthropic's Claude Code momentum among developers. The company is rolling out banked rate-limit resets for its Codex coding agent, letting paying users bank the resets they earn and spend them on demand rather than racing a fixed weekly clock.
Under the new policy, Go, Plus, Pro, and Business subscribers each get a single free banked reset to start. Plus and Pro users can earn a second reset by inviting up to three friends to try Codex over the next two weeks; once an invited friend sends their first Codex message, both the inviter and the invitee receive the bonus reset, per The Decoder's reporting on the rollout. OpenAI frames the change as a response to developer demand for more flexibility, not as a price reduction.
The move lands in a pricing environment that OpenAI's own chief executive has publicly flagged as a problem. At an OpenAI enterprise event on June 4, 2026, CEO Sam Altman called AI costs "a huge issue" for businesses and "probably the second biggest theme" of the year after agent adoption, citing Uber burning through its full-year AI budget in a single quarter as evidence. The banked-reset feature is a way to acknowledge that complaint without committing to a token-price cut that would show up directly on the S-1 OpenAI confidentially filed with the SEC on June 8, as The Decoder reports.
The competitive pressure is real. Anthropic's Claude Code has gone viral among developers, and the company recently passed OpenAI in valuation for the first time. The Wall Street Journal, as relayed by The Decoder, says OpenAI is weighing token-price cuts aimed at luring Anthropic customers and expects Anthropic to respond in kind. Neither company has moved on list prices yet, and the banked-reset rollout is the more visible of OpenAI's two current levers.
The economics of the change matter as much as the marketing. A banked reset does not add capacity; it lets heavy users shift when they consume what they would have consumed anyway. For developers, that means fewer all-nighters lost to a reset window that lands mid-build, and more ability to cluster heavy usage around the moments that matter, like a release sprint. For OpenAI, the cost is mostly a scheduling change rather than a revenue cut, and it lands before the price-cut conversation hardens into a number that investors would discount into the S-1.
There is friction in the design. Banked resets can be gamed: a user who knows they will hit a wall can hoard free windows and dump heavy workloads into the banked slots, smoothing their own consumption while leaving lighter users on the same fixed clock. The friend-invite bonus extends the same logic to distribution, trading one extra reset for a Codex install in someone else's workflow and expanding the product's footprint in exchange for it. Both moves privilege the heaviest users, who are also the loudest.
Context shapes the timing. Usage-based billing has largely replaced flat-rate enterprise plans at OpenAI, and The Decoder notes that the same workload that cost roughly $200 a month under flat-rate can scale to several thousand or tens of thousands on usage-based pricing, with some enterprise customers reportedly pulling back on AI spend. Banked resets do not lower that ceiling; they push some of the unpredictability onto the developer instead of the bill.
What to watch next: whether OpenAI follows the banked resets with the actual token-price cuts the Wall Street Journal says it is weighing, and whether Anthropic, which has also filed to go public later this year, responds with its own packaging or pricing move first. The next step will set the terms of the pricing fight that both companies' filings will eventually have to explain.