OpenAI Codex Head Tibo: Active Users Exceed 25 Million
Tibo, head of OpenAI Codex and ChatGPT, announced that Codex has reached 25 million active users, prompting a reset of usage limits for all paid subscriptions of ChatGPT Work and Codex.
He added that he originally intended to announce this milestone the day before, but the reset button had already been pressed once that day, causing the celebration to be postponed. The reset applies to all paid tiers, rather than just providing additional limits to a specific tier of users.
He clarified the package multipliers: although both ChatGPT and Codex are labeled as 20X, in Codex, 20X only applies to the weekly usage cap; both Pro tiers do not have a 5-hour limit. Pro 20X was described as "exactly 20 times the Plus usage," reflecting the literal multiplier.
In the same communication, he humorously referred to the company as The Reset Company and previewed more news to come. The reset is seen as part of the product rhythm rather than a one-time customer service compensation.
Public records show that this figure is a new milestone following a continuous increase: around 1 million weekly active users at the beginning of the year, about 3 million in April, approximately 5 million by the end of May, jumping between 6 million to 8 million in mid-July, reaching 20 million on August 21, and now 25 million.
In market mechanics, the payers are the paid developers and enterprise seats running continuous agent tasks, while the sellers are the limited-capacity windows. The event is driven by the user milestone, with funds temporarily shifting from "waiting or purchasing points after limits are exhausted" back to "concentrated consumption after free resets"; benefiting are paid subscription retention and peak usage, while pressure is on capacity scheduling and the gray usage of turning subscriptions into interface resales.
Recently, the official team has also addressed issues such as abnormal limit consumption, long conversation image compression, excessive deductions for conversation title features, and risk control flags for converting subscriptions into interfaces for redistribution, often using full resets to offset complaints.
Source: Public Information
ABAB AI Insight
Tibo transformed Codex from a research preview into a key agent in the ChatGPT desktop version, using growth strategies that do not involve price cuts but instead make limits a publicly pressable button. From resetting every time 1 million users are added, to continuing resets at 20 million and 25 million, the product head himself has become the quota switch, with the community even starting to track the number of resets. This turns customer service compensation into a retention ritual.
The capital path involves pre-collecting subscription fees and controlling marginal capacity through limits. Plus attracts customers with low prices, while Pro 5X and Pro 20X sell weekly limits as multiplier products, with point purchases serving as a second layer of monetization after limits are filled. The reset appears to be a concession, but it actually reignites the already pre-collected monthly fees into token consumption, changing "leave when used up" to "renew after another round of usage".
In comparison, Cursor charges per request, Claude Code limits by seats and windows, and GitHub Copilot initially captures editors with low prices. Codex is in an expansion phase transitioning from a coding tool to a general work agent, with ChatGPT Work users at one point announced to exceed Codex, indicating that the battlefield has expanded from coding to office agents, but the billing core remains the same weekly limit.
Structurally, this represents a transfer of pricing power. The mechanism is: as the capability gap of models narrows, whoever can price and publicly adjust "how many agents can run weekly" decides where developers lock their workflows. Limits create scarcity, resets create peaks, and multiplier packages mark scarcity as a price list.
ABAB News · Cognitive Laws
Limits create thirst, resets create loyalty
Prepaid sales are for limits, not models
Buttons retain paying users better than ads