OpenAI Codex Head Tibo: Cycle Reduced to Hours with One Billion Users
OpenAI's Thibault Sottiaux, responsible for Codex and ChatGPT core products, stated that scaling to around one billion users is more challenging, but the team is working to reduce related cycles to just hours or even minutes in the future.
He is publicly known as Tibo, serving as the external face of the coding agent Codex and overseeing core products and platforms. The community refers to his announced quota resets on social media as the "Tibo Reset": during milestones, fault compensations, or model updates, he replenishes or pre-stores quotas for paid tiers like Plus, Pro, and Business. He mentioned that he receives an average of one private message or email requesting a reset every six minutes; he also noted that about 5% of production traffic goes through Pi, and another 5% through OpenCode, which can be accessed using a ChatGPT account on third-party clients.
Active users of Codex grew from around one million in February 2026 to about five million weekly active users by early June, reaching a total of eight million with ChatGPT Work by mid-July, over 15 million by mid-August, and claiming 20 million at a certain point later that month. Initially, he promised to reset quotas every time the user count increased by one million until reaching ten million, but after surpassing that threshold, it changed to irregular pre-stored resets. During this period, there were temporary cancellations of five-hour window limits, rollback of measurements, fixes for cache hit rates, and instances where subscriptions converted to interfaces were flagged by risk control. On the ChatGPT side, he mentioned that voice capabilities have been rolled out to about one billion users.
The one billion refers to the scale of ChatGPT, with coding tools being a high-consumption subset within it. Reducing the time to hours or minutes targets the latency for quota recovery, model switching, or fault compensation, rather than eliminating the limits themselves. API prices were previously announced to be permanently reduced by 50%. Developers following this account have developed a countdown culture, with some sites creating reset radars based on his posts.
In market mechanics, the buyers are paid developers who need to continue generating code even after exhausting their quotas, while the sellers are OpenAI product leaders who maintain retention through public resets. Computational power is diverted from training and inference budgets to provide compensatory quotas, with social attention focused on the next button press. Events are driven by responses from product leaders. Beneficiaries are subscription users who can continue working during quota gaps; the pressured parties are cluster scheduling and anti-cheat systems, as well as gray market channels that resell subscriptions as interfaces.
He emphasized that changes in quotas will be explained to the community in advance and will not be silently altered.
Source: Public Information
ABAB AI Insight
Tibo has transformed quota policies that should have been documented on the status page into immediate powers of a personal account. Resets have shifted from operational actions to growth rituals: each time the user count crosses a threshold, a token is issued. This positions the product leader to simultaneously handle capacity planning, customer service, and community operations, aligning expectations for the billion-user-level ChatGPT with the twenty-million-level Codex through the same voice.
The capital implication is using short-term computational power to enhance retention. Free or pre-stored resets reduce the immediate friction of switching to other coding models like Claude, at the cost of sudden loads on training clusters. Compressing cycles from days to hours or minutes means requiring billing, risk control, and model routing to become near real-time systems, rather than weekly configuration updates. Third-party clients capture about ten percent of the traffic, indicating that while the official product is competing for entry points, it is also creating pipelines for the account system.
Similar structures appear in gaming operations with "compensation draws," service quotas after cloud vendor failures, and early Twitter announcements of feature toggles from employee accounts. The coding agent is in an expansion phase where limits are treated as product features: limits create scarcity, resets create events, and together they maintain daily active users.
Structural changes belong to technological substitution. Manual single-instance resets will be replaced by automated short-cycle distributions, as a billion accounts can no longer rely on a single person's sleep schedule to determine capacity. The mechanism is that generative usage has no natural upper limit and can only be segmented by time windows; the shorter the window, the closer it approaches per-second billing in the cloud, moving away from "subscription packages used to the limit and then waiting" consumption applications.
ABAB News · Cognitive Laws
- Limits create events, resets turn operations into growth.
- After surpassing one hundred million users, hourly releases become more challenging than the features themselves.
- The community revolves around one person's button, indicating that quotas have not yet been fully productized.