I am writing to express my concern about OpenAI’s announced December 2026 plan to merge ChatGPT and Codex usage allowances for Plus, Pro and Business subscribers.
I understand the motivation behind integrating agentic capabilities into the main ChatGPT experience and making resource allocation more flexible. I am not opposed to agentic features, nor am I demanding unlimited usage. However, I strongly believe that ordinary conversational usage and agentic execution should retain separate allowances.
These are fundamentally different activities, and combining them into a single pool creates several problems.
1. ChatGPT and Codex serve different purposes
Ordinary ChatGPT conversations involve generating responses, reasoning, discussing ideas, tutoring, and searching the Internet. Codex and agentic Work, by contrast, perform more autonomous and often resource-intensive tasks, such as coding, modifying files, operating applications, and executing multi-step workflows.
These activities complement one another, but they are not interchangeable. Users should be able to use both without constantly calculating whether one activity will deprive them of capacity for the other.
2. A shared pool risks becoming a disguised reduction in effective usage
Even if the nominal combined allowance appears reasonable, a shared pool can still disadvantage users who rely heavily on both products.
A user who previously had sufficient independent allowances for ChatGPT and Codex might find that intensive use of either product exhausts the capacity needed for the other.
The problem is therefore not exclusively the total number of tokens provided. It is also the loss of independently available capacity.
OpenAI should not silently reduce the effective value of an existing paid subscription through quota consolidation.
3. Conversational continuity and advanced reasoning matter
Many users rely on ChatGPT for extended discussions, academic work, technical reasoning, research, and tutoring.
These activities often require sustained access to more capable reasoning models. Instant chat, while useful, is not an equivalent replacement for higher-thinking capabilities.
Users should not have their ability to continue a demanding reasoning conversation compromised because they previously used Codex or delegated an agentic task.
4. The two types of usage have different consumption patterns
Agentic tasks can involve substantial context, multiple tools, repeated operations, and long-running execution. Their consumption can be concentrated and difficult for users to predict.
Ordinary conversations have different usage patterns.
Combining both into one token pool makes resource budgeting more difficult, particularly when consumption varies according to the model, reasoning effort, task complexity, and tools involved.
5. A shared allowance could discourage adoption of agentic features
Users may become reluctant to experiment with Codex or delegate tasks to Work if doing so consumes resources needed for their ordinary ChatGPT usage.
This creates an unfortunate incentive: users become more conservative about using the very capabilities OpenAI wants them to adopt.
A good quota system should encourage users to explore both products rather than make them compete against each other.
6. Transparency is essential
Users need to understand exactly what consumes their allowances.
They should be able to see separate usage balances, reset times, and clear explanations of how different models and activities are charged.
In particular, automatic tool invocation must not unexpectedly consume agentic allowances without the user’s knowledge.
7. Plus subscribers deserve predictable value
A $20 subscription does not imply unlimited computing resources. Reasonable usage limits are entirely understandable.
However, paying subscribers should receive a meaningful and predictable level of access to the core conversational and agentic capabilities included in their plan.
The value of one capability should not depend on how much of another capability the user has consumed.
8. There is a better alternative: separate allowances based on the nature of the activity
I would strongly prefer OpenAI to retain two independent allowances:
Conversational allowance
-
Ordinary ChatGPT conversations and responses.
-
Reasoning and discussions at all supported thinking levels.
-
Tutoring, technical discussions, and research.
-
Ordinary Internet searches and information retrieval.
Agentic allowance
-
Autonomous coding and debugging.
-
Multi-step delegated tasks.
-
Browser and computer-use automation.
-
File and application operations.
-
Agentic Work and Codex execution.
Importantly, usage should be classified according to the activity being performed, not the interface through which it was initiated.
For example, if I ask ChatGPT to perform an autonomous, multi-step operation, that operation should consume agentic allowance even though I initiated it through ChatGPT. Conversely, an ordinary conversation involving Internet searches should remain within the conversational allowance.
Where a conversation involves both activities, the agentic portion should be clearly identified and accounted for separately.
Users should also receive explicit disclosure before an operation consumes agentic allowance, together with separate usage meters and reset information.
If OpenAI considers a fully unified allowance unavoidable, I would prefer protected minimum allocations for both activities, with optional, user-directed transfers of unused capacity, rather than automatically combining everything into one pool.
Conclusion
I am not asking OpenAI to provide unlimited usage or abandon its agentic capabilities. I am asking it to preserve the distinction between two valuable but fundamentally different forms of usage.
ChatGPT helps users think, learn, discuss, and research. Codex and agentic Work help users execute and accomplish tasks.
These capabilities should complement one another, not compete for the same limited allowance.
Please reconsider the planned consolidation and retain separate conversational and agentic usage allowances. At the very least, publish the exact effective allowances, consumption rules, and safeguards before implementing the change.
Thank you for considering this feedback.