OpenAI Stabilizes Long-Running Codex Sessions and Resets User Limits

OpenAIOpenAI

· Updated

OpenAI resolved a technical issue in Codex that caused user usage limits to deplete faster than intended during long-running sessions. The team rolled back a faulty compaction optimization and performed a full reset of usage limits for all accounts to restore developer access.

OpenAI fixed a bug in Codex, its agentic coding platform, that caused user usage limits to drain prematurely. The engineering team traced the issue to a recent optimization for compaction—the process of condensing conversation history to fit within a model's context window.
Platform
Codex
Root cause
Compaction cache hit rate regression
Resolution
Optimization rollback and fix
Account impact
Full usage limit reset for all users

The faulty optimization negatively impacted cache hit rates during long-running autonomous sessions, forcing inefficient data re-processing. This fix follows updates aimed at enabling multi-day tasks, including the introduction of OpenAI's manual context compaction to help developers manage token footprints during complex workflows.

OpenAI has rolled back the change and performed a manual reset of usage limits for all accounts to compensate for the lost tokens. This restoration repeats a similar Codex limit reset performed earlier this month following performance degradations, ensuring developers can resume high-volume agentic work without waiting for their standard refresh cycle.

Tibo
Tibo
@thsottiaux
X

Some of you noticed limits drained faster in Codex, we root caused it to an optimization that we rolled back that had an impact on cache hit rates when compacting across long running sessions. We fixed this and have now reset usage limits for all accounts. Enjoy the weekend.

487retweets10klikes
View on X

Still wondering? A few quick answers below.

OpenAI identified a technical bug in Codex related to a recent optimization for long-running sessions. This change negatively impacted cache hit rates during context compaction, which is the process of condensing conversation history to fit within the model's memory. This caused the system to re-process data inefficiently, consuming user rate limits much faster than intended.

The engineering team at OpenAI root-caused the problem to a specific optimization that affected how the system handles long sessions. To resolve the issue, they rolled back the faulty optimization and implemented a fix to stabilize cache performance. This ensures that the compaction process now operates with the expected efficiency during autonomous engineering tasks.

OpenAI performed a full manual reset of usage limits for all Codex accounts following the fix. This action was taken to compensate users whose quotas were prematurely depleted by the technical error. All developers now have a fresh usage allowance to continue their work without waiting for their standard refresh cycle.

Context compaction is a technical process used by the Codex platform to manage long-running agentic sessions. It involves condensing the current state and conversation history into a smaller token footprint so the AI can continue working without exceeding its maximum context window. A recent failure in this process led to the rapid drainage of user limits.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

OpenAI

See all AI news & updates from OpenAI →

Share this update