# Codex Usage Reset August 30, 2026: Tibo on Longer Limits

> The August 30, 2026 Codex usage reset comes with token fixes from OpenAI. Tibo says the same limits may last 10% to 50% longer.

- Author: Kai Wang (AI Kai)
- Published: 2026-08-30
- Updated: 2026-08-30
- Topic: Codex
- Tags: codex, codex-usage-reset, codex-reset, token-usage, tibo-sottiaux
- Canonical: https://hqman.me/blog/codex-usage-reset-token-optimization/

Codex usage reset again on August 30, 2026. The familiar reset rhythm is back, and OpenAI says the same paid limits should now last longer after a round of token usage fixes.

Tibo Sottiaux says users may get 10% to 50% more usage from the same limits, depending on how they use Codex. This is not a new quota tier. It is less wasted usage inside the existing one.

## What OpenAI fixed

The team went through thousands of reports and found several ways Codex could consume usage without doing useful work.

### Compaction kept old images

When Codex compacted a long conversation, it sometimes kept images that no longer belonged in the active context. The larger context could trigger another compaction, creating more work. OpenAI says heavy image users saw about 10% extra usage from this problem.

### Memory workers and goals kept running

Background memory workers could inherit stop hooks and keep trying after they should have stopped. OpenAI saw one thread check whether it could stop 15,000 times.

The same pattern affected `/goal`. A goal could finish and continue past its stop condition, or Codex could keep retrying a broken tool. Some examples used 15% to 70% of a weekly allowance.

### Automations ran too often

Some custom automation schedules ran more often than users configured. The team fixed the scheduling issue, so a recurring task should follow its configured frequency.

### Subagents used stronger models than requested

Smaller models, including Luna, could select more capable helpers without an explicit request. Codex could also ask a subagent to run in `/fast` when the main model was not using that mode.

The fix keeps the model and speed choices closer to what the user requested. That should reduce surprise usage in multi-agent workflows.

### Computer History summarized overlapping activity

The older Computer History implementation could summarize overlapping activity more than once. OpenAI saw cases where this used up to 20% of weekly usage.

### Rolling summaries added background requests

Normal turns could trigger extra background requests for rolling task summaries. Each request was small, but together they added about 1% to token usage. OpenAI disabled this behavior.

### MCP results and instructions were processed twice

Some MCP tool results were encoded twice. In other cases, tool instructions were cut off and fetched again. Both paths added work without adding value, and the team fixed them.

## What this means for Codex users

The practical result is simple: your quota should feel more durable, especially if your workflow uses images, `/goal`, automations, Computer History, MCP, or multiple agents.

The 10% to 50% range is not a guaranteed increase for every account. It depends on which bugs affected your sessions. A user who never uses images or background agents may see little change. A heavy user who hit repeated compactions or runaway goals may see a much larger difference.

OpenAI is also working on showing usage details inside the app. That will make it easier to see which part of a workflow consumes the quota instead of guessing from the meter.

## The reset rhythm is back

OpenAI reset usage for paid Codex and ChatGPT Work users as the fixes rolled out. For Codex users, the important combination is the reset itself and a less wasteful usage curve afterward.

Thanks to Tibo and the Codex team for reading the reports, finding obscure failure modes, and fixing the systems behind them. The next useful test is simple: run the same kind of work you ran before the update and compare how long the quota lasts.

## Sources

- [Tibo Sottiaux on X](https://x.com/thsottiaux)
- [Earlier Codex usage reset and rate-limit fixes](https://hqman.me/blog/codex-usage-reset-rate-limit-fixes/)
- [The story behind Codex Reset, according to Tibo](https://hqman.me/blog/codex-reset-tibo/)