Mystery Solved: OpenAI Identifies Three Technical Causes of Excessive Codex Credit Use

Following user reports of unexpectedly rapid Codex credit depletion, OpenAI has concluded a detailed technical investigation. Lead Tibo Sottiaux confirmed the team pinpointed three specific internal processes responsible for the extra consumption, moving beyond earlier external explanations.

The Three Culprits Behind the Drain

The investigation revealed the primary issues stem from these core functionalities:

  • Inefficient Context Compression in Image-Heavy Sessions: To handle long tasks, Codex continually compresses older conversation content. However, the current process becomes notably inefficient during repeated compression cycles in sessions containing many images, leading to unplanned credit usage.
  • Usage Spikes from the "Computer History" Feature: This recently launched feature, which brings Mac app and browser activity into conversations, was found to consume more resources than anticipated in certain high-demand scenarios.
  • Hidden Costs of Auto-Title Generation: Even the auxiliary function that automatically generates conversation titles was using more computational resources than the initial models projected.

Sottiaux had previously noted no systemic anomalies and pointed to factors like API calls and account sharing. As the probe deepened, he acknowledged that reduced cache hit rates for some users might have accelerated credit burn. This latest update shifts the focus squarely onto optimizing Codex's own workflows.

The Fix and User Compensation Plan

OpenAI has developed a patch scheduled for release this Sunday (US time).

Critically, the company announced it will perform a full credit reset for all paid subscribers' Codex usage. This measure is intended to compensate users for credits lost due to the identified technical issues. The reset is expected around 2 PM Pacific Time on Sunday.

Looking Ahead: Ongoing Efficiency Gains

Beyond addressing the immediate problems, the investigation led the team to an unrelated new optimization that reportedly can significantly improve Codex's efficiency. This enhancement is slated for deployment next week, indicating a push to improve the system's fundamental resource economics.

From problem identification to providing a concrete fix and compensation, this response highlights OpenAI's attention to developer experience regarding resource consumption. The credit reset and forthcoming optimizations should help restore confidence in the predictability and control of Codex service costs.