The opacity of large language model usage limits has finally been pierced by a new community tracker, usageupdates.com, which logged 76 distinct usage resets and limit changes in a mere 5.0-day average window. The platform, which appeared on Hacker News this week, reveals that OpenAI and Anthropic are engaged in a frantic cycle of quota resets, credit grants, and limit extensions to manage the surge in developer demand. With 124 total tracked changes as of early October 2026, the data exposes the volatile infrastructure behind the AI race, where 'unlimited' usage is often a marketing fiction patched together by manual server-side overrides.

The OpenAI Reset Cycle

OpenAI’s activity has been particularly chaotic, with a worldwide Codex and ChatGPT Work usage reset triggered just two days ago during their DevDay conference. This follows a pattern of frequent interventions, including a one-time $2,500 usage credit granted to Pro 200 subscribers and a reset for Plus, Pro, and Business users nine days ago. The tracker highlights a critical vulnerability in OpenAI’s 'banked reset' system, noting an incident 22 days ago where OpenAI had to grant replacement resets after the initial banked versions failed to apply. This technical debt is compounded by recent 'usage-drain bugs' identified 39 days ago, forcing yet another quiet reset of Codex and ChatGPT Work limits.

Anthropic’s Strategic Boosts

Anthropic’s strategy, while less frequent, involves significant one-time boosts tied to model launches. Eight days ago, the company issued one-time credits of $100 for Pro and $250 for Max users specifically for Claude Code cloud sessions. This coincided with a 20% raise in Claude Code 5-hour limits and the introduction of a bankable reset at the Opus 5.5 launch. Earlier in September, Anthropic reset weekly limits for the Fable 5.1 launch weekend, a move that appears to be part of a broader effort to stabilize user experience during major version updates. The data shows Anthropic’s limits were kept 50% higher through September 13, indicating a sustained push to retain enterprise developers.

Key Takeaways

  • OpenAI triggered a global usage reset during DevDay, signaling that launch events are now critical pressure points for infrastructure stability.
  • Anthropic is using targeted credits ($100-$250) to subsidize cloud session costs, shifting from pure rate-limit increases to financial compensation.
  • The 'banked reset' feature introduced by OpenAI has suffered from implementation bugs, requiring manual intervention and replacement credits.
  • Usage limits are no longer static; they are dynamic tools used for marketing (launches) and crisis management (outages and bugs).

The Bottom Line

The era of predictable AI usage pricing is dead; we are now in an era of 'reset-as-a-service,' where trust is maintained not by stable limits, but by the frequency of apologies and credit grants.

Infrastructure Reality Check

The sheer volume of resets—73 days since the longest gap—suggests that the underlying inference infrastructure is struggling to keep pace with the 'prosumer' developer market. OpenAI’s recent move to trim Astra’s usage draw by up to 4x for power users is a clear indication that token economics are being squeezed. Developers must now treat usage limits as a variable cost subject to real-time policy changes, rather than a fixed contract. The tracker’s existence alone is a indictment of the current lack of transparency, forcing the community to build its own observability layer for vendor behavior.