Subscription-based coding agents like Claude Code and Codex enforce 5-hour and weekly LLM usage quotas. Recently, Anthropic and OpenAI have reset weekly quotas far more frequently, often without official announcement. OpenAI reset the Codex weekly quota six times in two weeks: July 9, July 10 (twice), July 14, July 15, and July 17, plus banked resets on July 12–13 tracked by codex-resets.com. These resets, worth roughly $25 on a $100/mo plan, arrive unpredictably—sometimes hours before the natural reset, wasting the gifted capacity.

As a $100/mo subscriber, I monitor usage closely to exhaust each week's quota. After GPT-5.6 Sol launched, nearly every reset hit when my quota was at 50%+, making me feel I "wasted" $12. The random timing forces abrupt workflow changes to burn quota before it resets again, turning a dopamine hit into a net deficit. Thibault Sottiaux even polled users on whether resets are excessive.

The surge coincides with July's flood of frontier models: Fable 5, GPT-5.6 Sol, Grok 4.5, Muse Spark 1.1, and Kimi K3. The cynical view: frequent resets prevent power users from testing competitors once their quota expires. If resets persist, they'll lose meaning—and I'll downgrade to the $20/mo plan to avoid waste, likely not OpenAI's goal.