Codex is burning weekly limits much faster: token efficiency regression makes 20x Pro feel insufficient #27242

Codex app version

Desktop Codex: 26.602.71036 — released Jun 8, 2026

Platform / plan

• macOS Tahoe 26.5.1
• Apple Silicon MacBook Pro
• Pro plan with 20x Codex usage limits

Summary

Recently, Codex appears to have become much less token-efficient. Completing the same kind of development task now seems to consume much more weekly limit usage than before.

From the user side, this feels like either a model-quality regression, a planning/implementation efficiency regression, or a routing/behavior change that causes Codex to spend many more tokens and much more usage to finish a task.

Observed behavior

• Recent Codex tasks consume noticeably more weekly limit usage than similar tasks used to.
• A task that previously felt reasonable now requires far more usage to complete.
• The effective usefulness of the 20x Pro weekly limit feels much lower than before.
• I used to be able to move several projects forward in parallel.
• Now the weekly limit feels much less durable, so I can realistically focus on only one project at a time.
• If I try to run multiple projects in parallel like before, my weekly limit can be exhausted within only two or three days after reset.

Expected behavior

Codex should remain reasonably token-efficient for normal coding workflows. A Pro user with 20x weekly limits should be able to make meaningful progress across multiple projects, especially if the same types of tasks were previously feasible within the weekly limit.

If the model or backend behavior changes in a way that significantly increases usage consumption per completed task, users should have visibility into what changed.

Actual behavior

The same general workload now appears to consume much more week-limit usage. This makes the 20x Pro limit feel insufficient for the workflow that previously worked.

Why this matters

Weekly usage limits are only useful if the model remains efficient enough to complete tasks within those limits. If Codex needs far more tokens/usage to finish ordinary tasks, then the practical value of the paid Pro limit drops significantly.

This affects real development workflows. I previously used Codex to advance multiple projects at the same time, but the recent usage efficiency regression forces me to reduce the workflow to one project or risk exhausting the weekly limit shortly after reset.

Request

Please investigate whether there has been a recent regression in Codex token efficiency, task execution efficiency, or GPT-5.5 behavior that causes tasks to consume much more weekly usage than before.

It would also help to provide:

• clearer usage accounting per task,
• visibility into why a task consumed a large amount of usage,
• improvements to reduce unnecessary token consumption,
• and guidance if recent backend/model changes intentionally changed the expected usage cost per task.

This is not only about asking for more quota. The core issue is that Codex seems to require much more usage to complete comparable tasks, which makes the weekly limit feel much less usable than before.