A Proposal for Fair, Predictable, and Transparent ChatGPT Work & Codex Usage Limits
Replace unpredictable resets with clear subscription allowances, fixed reset schedules, and optional on-demand capacity.
OpenAI’s recent experiments with usage limits, global resets, and banked resets have highlighted a fundamental problem: users need predictable access to computing resources, not occasional generosity.
Frequent free resets may be appreciated, but they are not a sustainable substitute for a transparent usage policy. Users should be able to understand what they are paying for, plan their workloads, and purchase additional capacity when necessary.
I propose a simpler and more consistent system.
1. Establish a Clear, Proportional Pricing Structure
| Monthly Subscription | Weekly Usage Allowance | Five-Hour Limit |
|---|---|---|
| $20 | 1× | Yes |
| $100 | 10× | No |
| $200 | 20× | No |
| $500 | 50× | No |
Under this structure, all Pro tiers provide the same amount of weekly capacity per dollar. Higher subscription tiers simply provide proportionally greater resources.
The $20 plan remains an accessible entry point, while the Pro tiers reward users who commit to larger subscriptions.
Most importantly, these multipliers must represent consistent, measurable computing capacity, rather than percentages whose underlying value can change without explanation.
Model-specific consumption rates should be clearly published, and any changes should be communicated in advance.
2. Keep Regular Weekly Reset Dates Fixed
A subscription’s regular weekly reset schedule should be predictable.
Additional resets, purchased capacity, promotional rewards, or compensation should never postpone the next scheduled weekly reset.
A user whose allowance normally renews on Friday should continue to receive that renewal on Friday, regardless of whether they purchased or redeemed additional capacity on Tuesday.
This distinction is essential: a regular subscription renewal is an entitlement, not a reward that can be delayed because the user received something extra.
3. Offer Purchasable Weekly Capacity Cards
Users occasionally need significantly more computing capacity than their normal subscription provides.
Instead of requiring them to upgrade their entire subscription for a temporary project, OpenAI should offer optional weekly capacity cards.
For example, a $200 subscriber with a 20× weekly allowance could purchase an additional 20× capacity card when needed.
These cards should follow several principles:
-
Additive capacity: Redeeming a card adds its full allowance without replacing or erasing unused subscription capacity.
-
Fixed renewal schedule: Using a card never changes the regular weekly reset date.
-
Separate balances: Purchased capacity is tracked independently from subscription allowances.
-
Persistent value: Purchased capacity survives regular weekly resets until consumed or until its clearly disclosed expiration date.
-
Subscription independence: Changing or downgrading a subscription does not confiscate previously purchased capacity.
-
Transparent pricing: The cost and capacity of each card are disclosed before purchase.
Subscription allowances should normally be consumed first, followed by purchased capacity, with clear exceptions when expiration dates make another order more beneficial to the user.
This would give users a practical way to handle temporary spikes in demand without disrupting their subscriptions.
4. Reserve Free Resets for Compensation
Global free resets should no longer be distributed unpredictably as a routine promotional mechanism.
Instead, complimentary capacity should primarily serve as compensation for verified service failures, incorrect usage deductions, prolonged outages, or other situations where users have lost resources through no action of their own.
Compensation should be proportional to the actual impact whenever reasonably measurable.
This would be fairer than resetting everyone indiscriminately. Under the current approach, someone who has exhausted their allowance may receive substantial additional capacity, while another user with 90% remaining receives very little practical benefit.
Compensation should also be additive and must never delay the normal subscription reset.
Promotional capacity, if offered, should likewise be clearly identified and managed separately from compensation.
5. Make Usage Accounting Auditable
None of these improvements will be sufficient if users cannot understand how their allowances are consumed.
OpenAI should provide:
-
Clear weekly capacity totals, not just remaining percentages.
-
Itemized usage records showing models, reasoning levels, speed modes, and agent activity.
-
Visibility into the consumption of parallel agents and delegated tasks.
-
Advance notice of changes to usage multipliers or subscription capacity.
-
A mechanism for disputing unexpected deductions.
-
Separate accounting for subscription, purchased, promotional, and compensation balances.
A 20× or 50× allowance must have a stable and understandable meaning. Otherwise, the advertised multiplier offers little practical certainty.
6. Why This Is Better for Both Users and OpenAI
For users, the benefits are straightforward: predictable renewals, transparent pricing, no unexpected loss of purchased capacity, and the freedom to acquire additional resources when necessary.
For OpenAI, the system could provide more predictable subscription commitments, a legitimate revenue stream for temporary high-demand usage, and fewer disputes over arbitrary resets or unclear allowances.
It would also reduce the pressure to distribute frequent global resets simply to maintain user satisfaction.
There are legitimate infrastructure and capacity-management considerations, but these should be addressed through transparent scheduling, optional priority mechanisms, and published usage rules rather than unpredictable changes to the value of a subscription.
Conclusion
The goal should not be to give users unlimited computing resources. It should be to give them exactly the resources they have paid for, under rules they can understand and rely upon.
The proposed structure is simple:
$20 = 1×, $100 = 10×, $200 = 20×, $500 = 50×.
Pro subscriptions should have no five-hour restrictions. Weekly renewal dates should remain fixed. Additional capacity should be available for purchase without interfering with regular renewals. Free resets should primarily compensate users for genuine service failures.
A predictable system would ultimately be more valuable than repeated surprise resets.
Users should not have to wait for someone to press a reset button to feel that their subscription is worth what they paid for.