OpenAI is continuing its unusual 28-day Codex improvement campaign, giving users another usage reset as the company balances new AI coding features with growing demand for its coding agent.
OpenAI has started October with a very different approach to improving Codex.
Instead of simply announcing a large collection of new features at once, the company is rolling out a daily sequence of improvements — or, on days without a major improvement, giving users a full usage reset.
The initiative has quickly become a major talking point among Codex users because usage limits have become increasingly important as developers rely on the coding agent for longer and more complex tasks.
OpenAI’s developer community says the company committed to 28 days of quality-of-life improvements or a full reset, with the goal of making Codex simpler, more efficient and more useful.
Codex users are getting another reset
The latest reset comes as part of that ongoing experiment.
The basic idea is straightforward: each day, OpenAI either ships an improvement that should benefit Codex or Work users, or provides a full reset when there isn’t a qualifying improvement.
That makes the initiative unusual compared with a conventional product update cycle.
Instead of waiting weeks for a large Codex release, users are effectively getting a stream of smaller changes, performance improvements and occasional opportunities to refresh their usage.
For developers who regularly run into Codex limits, the resets can be particularly valuable.
OpenAI’s current documentation confirms that eligible users can receive automatic one-time resets that refresh covered Codex or ChatGPT Work usage limits. However, eligibility and the specific limits affected can vary by offer.
The 28-day Codex experiment began with faster models
The first improvement in the campaign focused on performance.
OpenAI said GPT-6 Astra and GPT-6.1 Sol should feel around 50% faster by default when used through supported subscriptions and partner applications.
The improvement doesn’t require users to change a setting, with OpenAI describing it as a default response-speed improvement.
That is particularly relevant to Codex because speed can have a major effect on how developers work.
When an AI coding agent can reason, edit files, run tools and iterate more quickly, developers can move through debugging and implementation cycles with less waiting.
But faster coding agents can also encourage heavier usage.
That makes the relationship between performance and usage limits increasingly important.
Auto-review is now part of the Codex experience
Another significant change in the campaign is Auto-review.
OpenAI has made Auto-review available without counting it against users’ plan usage limits when they are signed in with their ChatGPT account.
Auto-review is designed to reduce the number of times Codex needs direct human approval while still maintaining oversight around potentially sensitive actions.
OpenAI’s research on the system says Auto-review can dramatically reduce the frequency of synchronous human approvals during coding-agent sessions while retaining an additional review layer for boundary-crossing actions.
For developers running longer autonomous coding tasks, that can make Codex feel considerably less interruptive.
Codex is becoming more agentic
The timing of these changes is important because OpenAI is simultaneously pushing its products toward more autonomous AI agents.
At DevDay 2026, OpenAI introduced Dots, described as always-on agents designed to handle ongoing responsibilities and work on behalf of users.
That broader move toward agentic AI also affects how Codex is positioned.
The coding assistant is no longer simply a chatbot that generates snippets.
Developers increasingly expect it to inspect repositories, modify files, execute commands, test implementations, review changes and continue working through multi-step tasks.
That makes reliability, speed and usage capacity just as important as the underlying model.
Usage limits remain a major concern
The excitement around the resets also highlights a problem that Codex users have been discussing for months: AI coding agents can consume usage surprisingly quickly.
A complicated coding task can involve far more model activity than a conventional chat request because the agent may repeatedly reason, inspect files, run tools, review its own work and make additional changes.
OpenAI’s documentation explains that Codex usage depends on factors including the model being used, task complexity, context, reasoning, speed and tools.
This means two developers can perform what appears to be a similar task while consuming very different amounts of their available allowance.
For power users, that makes every reset particularly noticeable.
OpenAI is also adding more workflow features
The 28-day campaign isn’t limited to usage resets and performance.
The Codex and Work ecosystem is also gaining broader workflow improvements.
OpenAI’s developer community has highlighted new functionality such as meeting notes, while users have also been tracking changes around planning and agent-based workflows.
Meanwhile, Dots represent OpenAI’s larger push toward AI systems that can remain active and take on ongoing responsibilities rather than waiting for every individual prompt.
The direction is clear: OpenAI wants its AI tools to do more of the work rather than simply help users perform individual steps.
What happens when you reach a Codex limit?
OpenAI’s current help documentation says users approaching or reaching a Codex limit can check their Settings or usage dashboard to see which allowance has been exhausted and when the relevant limit resets.
In an active Codex CLI session, users can also enter /status to check usage information.
If a user reaches a limit during an active turn, Codex can continue working on that turn under applicable fair-use rules. Afterward, the available options can include adding credits, using an available reset, upgrading or waiting for the normal reset.
OpenAI also makes an important distinction between automatic resets and banked resets.
An automatic reset is applied directly to the eligible allowance, while a banked reset can be saved and manually applied before its expiration.
The bigger story is OpenAI’s response to Codex demand
The repeated resets are interesting because they show how quickly AI coding has moved from an experimental feature into a serious development workflow.
Developers are increasingly using AI agents for real repositories and lengthy tasks.
That creates a difficult balance for OpenAI.
The company needs Codex to be powerful enough to handle meaningful software projects, fast enough to feel useful and autonomous enough to reduce manual work.
At the same time, those capabilities require substantial compute.
Usage limits therefore become a central part of the product experience.
OpenAI’s 28-day campaign appears to be an attempt to address several sides of that equation at once: improve performance, remove unnecessary friction, add new capabilities and periodically give users additional room to work.
Codex could look very different by the end of the 28 days
If OpenAI maintains the current pace, Codex users could see a surprisingly large number of changes before the experiment ends.
The company has already moved from faster model performance to free Auto-review, while additional workflow and agent features are appearing across the wider ChatGPT and Codex ecosystem.
For developers, the most interesting question may therefore not be whether another reset arrives.
It is what Codex looks like after OpenAI has spent 28 consecutive days removing friction from the coding workflow.
If the company can combine faster responses, better autonomous execution, fewer approval interruptions and more generous opportunities to recover from usage limits, Codex could become considerably more practical for sustained software development.
For now, another reset gives heavy users some breathing room — while OpenAI continues its unusually aggressive push to improve the coding agent day after day.
For continuing updates on OpenAI, ChatGPT, Codex, GPT models, and AI industry developments, bookmark:
