OpenAI to merge ChatGPT and Codex usage limits in December

The move to a shared pool

OpenAI has announced that starting in December, the usage allowances for Plus, Pro, and Business plans will be shared across both ChatGPT and Codex. This change follows an existing structure where the 'Work' and Codex products already share a single usage pool. The company is expanding this unified meter to include the main ChatGPT experience.

The primary goal of this consolidation is to bring the agentic capabilities of Work into the main ChatGPT interface. By merging these limits, OpenAI is creating a more cohesive environment where different types of AI tasks draw from the same resource. This allows the company to manage how users interact with various tools under a single subscription umbrella.

Users will be able to see their combined remaining usage and their specific reset times within their account settings. When a user reaches their limit, they will have the option to wait for the allowance to reset, upgrade their current plan, or purchase additional credits. This ensures that high-intensity users have clear paths to continue their work if they exhaust their limits quickly.

How the token system works

The new shared allowance is based on tokens, which are the fundamental units of information that a model processes and generates. Instead of a fixed number of messages, the amount of allowance a specific request uses depends on the model selected and the complexity of the task. This means that simple queries will cost less than more intensive operations.

OpenAI notes that more advanced models and complex tasks generally use more of the user's allowance. For example, a request involving a highly sophisticated model will deplete the shared pool faster than a request to a smaller, faster model. This system allows OpenAI to price and manage their computing resources based on the actual demand of the user's specific prompts.

One notable exception to this new shared meter is Instant chat. OpenAI stated that Instant chat will not draw from the usage allowance, meaning it remains outside the token-based tracking system. This distinction ensures that basic, immediate interactions do not impact the budget reserved for more complex tasks in ChatGPT and Codex.

Current usage estimates

While the December change introduces a new shared pool, OpenAI has provided estimates for the current Work and Codex allowance on Plus plans. These figures are not guaranteed message counts but serve as a guide for what users can expect in a five-hour window. The estimates vary significantly depending on which specific model is being utilized for the task.

For instance, OpenAI estimates that users might get between 5 and 45 local messages using GPT-6 Astra in a five-hour period. In contrast, using GPT-6 Luna could allow for between 350 and 3,000 messages in that same timeframe. This highlights the massive difference in consumption rates between different model tiers and their respective capabilities.

By performing arithmetic on these published ranges, it is clear that the difference in capacity is substantial. At the ends of the published ranges, a user could perform about 67 to 70 times as many messages with GPT-6 Luna as they could with GPT-6 Astra. These numbers provide a snapshot of the current balance before the December merger takes effect.

Impact on different users

For developers who rely heavily on Codex, this change introduces a new trade-off. A heavy coding session using a high-powered model like Astra will use up the allowance that would otherwise be available for standard chatting in ChatGPT. This creates a 'one meter' environment where different types of work compete for the same resources.

Business and Pro users will also see their workflows affected by this unified budget. Because the allowance does not correspond to a fixed number of messages, users will need to be more mindful of which models they choose for specific tasks. Choosing a more advanced model for a simple task may result in reaching the usage limit much faster than anticipated.

The company expects that most users will not need to wait for their usage to reset more often than they do currently. However, for power users who push the boundaries of the tools, the shared pool may require more strategic planning. This shift represents a move toward rationing agentic work inside a subscription model based on actual computational demand.

The broader AI landscape

OpenAI's move toward shared usage pools is not happening in a vacuum. Other major AI providers are also making adjustments to how they manage limits for their high-end plans. For example, Anthropic recently announced changes to its Claude Ultra plan, which is part of its $200 tier.

Anthropic is removing the five-hour limit for verified Ultra accounts, though their weekly limits remain unchanged. These concurrent shifts in the industry suggest a trend toward refining how premium users access high-performance models. Providers are looking for ways to balance high demand with the costs of running advanced AI systems.

As these companies refine their subscription models, the focus is shifting toward how agentic work is rationed. By using a single meter across chat and coding, OpenAI is setting specific terms for how users can utilize their paid plans. This evolution will likely continue as models become more capable and the tasks they perform become more complex.