Users Claim OpenAI Secretly Cut Codex Usage Limits by Half

Some OpenAI Codex users are raising concerns about their weekly usage allowances being consumed more quickly than expected. There is no proof that OpenAI has secretly cut Codex allowances, but developers want clearer information about how coding tasks affect their limits. A desktop app called NerfTrack, created by developer Ayaan Lashari, estimates API-equivalent weekly values of Codex usage.

OpenAI provides a token-based rate card explaining how credit consumption occurs based on input, cached input, and output tokens. The company also explains that more complex tasks can deplete limits faster. However, OpenAI does not provide a single weekly allowance that applies to every user and workload. Individual user reports on GitHub and OpenAI's Developer Community forum show that some users find it difficult to predict how much work their subscriptions will cover.

One user reported that their weekly allowance fell from 100% to 91% during a single 4.5-hour session. Another user noted that a single large prompt consumed around 50% of their five-hour allowance. Despite upgrading to higher plans, some users have reported exhausting their weekly allowances and additional credits sooner than expected.

OpenAI states that the lack of a simple task-by-task breakdown has become one of the main complaints among developers.