# ChatGPT and Codex Shared Limits: What Changes in December 2026

ChatGPT and Codex will share usage limits from December 2026. Learn what stays unlimited, how Pro is affected, and which details remain unconfirmed.

Canonical URL: https://aiidelist.com/blog/chatgpt-codex-shared-usage-limits-december-2026

Language: en

Published: 2026-11-01

Updated: 2026-11-01

## Key Takeaways

- **OpenAI has announced shared ChatGPT and Codex allowances starting in December 2026. Instant chat is excluded.** 
- **Work and Codex already share usage.** Their current limits depend on the plan, model, task, and settings. 
- **A Pro subscription does not provide unlimited GPT-6 Pro usage.** The subscription name and the model’s allowance are different things. 
- **The practical question is completed work per subscription dollar.** A shared allowance alone does not establish whether an individual subscriber will receive better or worse value.

**Policy checked: November 1, 2026.** Current limits below describe the documentation available on that date; they should not be treated as confirmed December entitlements.

## What OpenAI Has Confirmed About December

OpenAI’s [usage-limit announcement](https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits) names **Plus, Pro, and Business**. It connects the change to bringing Work’s agent capabilities into the main ChatGPT experience.

Usage will be measured in tokens, with consumption depending on the model and task. Combined remaining usage and reset times will appear under **Settings → Usage**.

The announcement does not specify a December launch day, numerical allowances by tier, or how existing model-specific caps will change. It also does not establish this transition’s rules for Free, Go, Enterprise, or Edu.

OpenAI expects most users will not need to wait for resets more frequently. That is a stated expectation, not a guaranteed outcome for every workload. 

## Is Unlimited ChatGPT Chat Ending?

**The current pricing page still lists everyday text chats as unlimited, subject to abuse guardrails.** It separately describes access to reasoning, research, uploads, and agent capabilities. The scope of the advertised feature matters. 

Three distinctions help explain apparently contradictory limit notices:

| Term | What it describes | What it does not establish |
| --- | --- | --- |
| Unlimited everyday text chats | The advertised allowance for ordinary text conversations | Unlimited access to every model or tool |
| Access to a model | Eligibility to select and use that model | An unlimited model allowance |
| Included Work and Codex usage | Agent activity covered by the subscription | A fixed number of completed projects |

A subscriber can therefore retain ordinary chat access while encountering a restriction on a more demanding capability. That is consistent with a product offering several different allowances.

Before interpreting a warning as the end of unlimited chat, identify **which model, feature, and usage window** the warning actually names.

## How Chat, Work, and Codex Limits Currently Differ

The current documentation distinguishes Chat model limits from the shared agent allowance:

| Experience | Current documented behavior |
| --- | --- |
| Chat with Thinking | Limits depend on the plan and model; a fallback may be available |
| GPT-6 Pro in Chat | Limited usage; a Pro subscription does not make this model unlimited |
| ChatGPT Work | Uses the allowance shared with Codex |
| Codex | Uses the allowance shared with Work |

OpenAI’s [Chat model documentation](https://help.openai.com/en/articles/20001354-gpt-6-and-other-models-in-chatgpt) explains the Chat restrictions, while its [Work and Codex usage guide](https://help.openai.com/en/articles/20001516-managing-usage-with-gpt-6-astra-in-work-and-codex) describes the agent allowance. 

**A model name alone is insufficient for understanding billing.** The surface, authentication method, selected settings, and account entitlement also matter.

For example, asking for a short explanation and asking an agent to investigate a repository are different workloads even if both begin with one message. Counting prompts hides the work performed after each prompt.

## Why One Message Can Consume More Than Another

OpenAI documents several drivers of Work and Codex consumption: model choice, input and output size, reasoning effort, speed settings, and tasks involving multiple steps. Increasing reasoning effort can consume more allowance without necessarily improving the result. 

Consider two illustrative requests:

- **Focused edit:** Rewrite a supplied paragraph to 100 words.
- **Open-ended investigation:** Research a market, compare competitors, inspect supporting material, and prepare a report.

Both are single user messages. The second can require substantially more material to be processed and more intermediate work.

This makes **messages per day a weak comparison metric** for agent subscriptions. A more useful measure is the allowance consumed to produce an acceptable result, including corrections and retries.

A cheaper model that repeatedly misses the requirement may be less economical for a particular task than a stronger model that completes it successfully. Conversely, selecting the strongest model for every formatting change may spend capacity without improving the deliverable.

## Tokens, Allowances, and Credits Are Different Quantities

OpenAI’s [pricing documentation](https://learn.chatgpt.com/docs/pricing) distinguishes tokens processed by models from credits used to pay for eligible activity. It also cautions against using API token prices to estimate included subscription tasks. 

| Quantity | Meaning | Practical consequence |
| --- | --- | --- |
| Tokens | Units of information processed or generated | Prompt length alone does not describe the entire workload |
| Included allowance | Usage covered by the subscription | Check the account’s remaining capacity and reset conditions |
| Usage credits | A balance for supported additional usage | Availability and eligible features depend on the account |

The [personal credits guide](https://help.openai.com/en/articles/12642688-using-credits-for-flexible-usage-in-chatgpt-personal-plans) says included usage is consumed first, followed by credits for supported activity. ChatGPT usage credits are not API credits. 

Avoid calculations such as dividing a subscription’s monthly price by an API output-token rate and presenting the result as a guaranteed token allowance. That calculation mixes different products and omits the subscription’s actual entitlement rules.

## What Pro 100, Pro 200, and Pro 500 Currently Include

OpenAI currently documents three monthly Pro tiers:

| Plan | Monthly price in USD | Relative included usage | Ultrafast access |
| --- | --- | --- | --- |
| Pro 100 | $100 | Lowest of the three Pro tiers | Not included |
| Pro 200 | $200 | More than Pro 100 | Not included |
| Pro 500 | $500 | Highest of the three Pro tiers | Included |

The [Pro tiers documentation](https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers) also confirms that the previously announced Pro 200 reduction was canceled. Existing subscribers with the higher Work and Codex allowance retain it; the lower allowance doubled to match. Active subscribers do not need to cancel and resubscribe. 

Separately, **Pro 100, Pro 200, and Pro 500 currently have no five-hour Work and Codex limit**, but they still have an included usage allowance. On plans where both five-hour and weekly limits apply, remaining capacity is required in both windows. 

Removing a short usage window and removing the overall allowance are different changes. A subscriber may be able to concentrate more work into one session while still exhausting the longer-period allowance.

## Faster Responses Can Use Allowance Faster

Speed settings deserve attention when comparing subscription value. The current Work and Codex pricing guide lists these multipliers relative to Standard speed for the same model:

| Speed | Included allowance consumption | Purchased-credit consumption |
| --- | --- | --- |
| Standard | Baseline | Baseline |
| Fast | 2.5× | 2× |
| Ultrafast for supported Astra and Sol models | 8× | 6× |

These are consumption multipliers, not promises of equivalent performance improvements. Eligibility restrictions still apply. 

An illustrative normalization shows the distinction: processing that costs 10 allowance units at Standard would cost 25 under a 2.5× multiplier, assuming otherwise identical processing. This is arithmetic, not an estimate of actual messages or tasks.

The useful decision is whether reduced waiting time justifies the additional consumption. Fast responses may be valuable during an interactive debugging session, while a background task with a flexible deadline may offer less reason to pay the speed premium.

## What Happens When an Allowance Runs Out?

**First identify the exhausted allowance.** OpenAI’s Codex guidance directs users to inspect the usage dashboard, credit balance, and reset time. In an active Codex CLI session, enter:

```text
/status
```

Starting a new conversation does not reset usage limits. 

Depending on account eligibility, the available responses can include waiting, using supported credits, changing plans, or applying a reset. Do not assume every option applies to every restriction.

**Credits:** These fund eligible additional usage after included capacity is exhausted. If automatic reload is enabled, check its spending controls. OpenAI notes that its maximum monthly reload setting applies to automatic reloads, not separate one-time purchases. 

**Purchased instant resets:** Eligible Plus and Pro personal accounts may be offered an immediate reset. It cannot be saved for later and changes the weekly schedule: the next weekly reset occurs seven days after the first Work or Codex request following the reset. Availability varies by account. 

Compare the displayed price with the time remaining until the next automatic reset and the work that must be completed. A purchase made minutes before a scheduled reset has a different practical value from one needed at the beginning of a busy week.

## Who Should Pay Closest Attention?

The following are workload scenarios, not measured forecasts of the rollout’s effects.

| User pattern | What to evaluate |
| --- | --- |
| Primarily everyday text conversations | Whether any warning actually concerns a tool or advanced model |
| Research and coding in the same workweek | Whether completing research leaves sufficient capacity for development |
| Occasional intensive project sessions | Whether limits interrupt a deadline despite low usage on other days |
| Frequent Fast or Ultrafast use | Whether faster responses justify the consumption premium |
| Business workspace member | Whether the restriction comes from personal capacity, workspace budget, or access controls |

A shared allowance could improve flexibility for someone whose workload shifts between activities. It could also create competition between activities for someone who heavily uses several of them.

**Neither outcome follows from pooling alone.** The actual allowance, consumption rates, reset rules, and workload determine the result.

## How to Evaluate the Change Without Guesswork

Create a small baseline using real work before changing subscription tiers:

1. **Choose three recurring tasks.** For example: a competitor brief, a contained code fix, and a document revision.
2. **Record the settings.** Note the plan, model, reasoning effort, speed, and execution surface.
3. **Record remaining usage before and after.** Avoid overlapping tasks when trying to attribute a change to one activity.
4. **Assess the result.** Count required corrections and whether the output met its acceptance criteria.
5. **Repeat comparable work after rollout.** Keep task scope and settings as consistent as possible.

Useful records include allowance change, elapsed time, additional spending, and successful completion. Dashboard percentages may be rounded, so several representative observations are more informative than one small task.

For developers, an acceptance criterion might be a reproduced bug fixed with the relevant checks passing. For researchers, it might be a complete comparison with verified links and no unsupported claims.

This approach evaluates **usable output**, rather than rewarding a configuration simply because it generates more messages.

## Frequently Asked Questions

**Does buying credits unlock every ChatGPT model or feature?**

No. OpenAI explicitly states that credits do not provide additional usage for every model or feature. Buying credits on Pro 100 or Pro 200 also does not currently unlock Ultrafast. 

**Can switching between Pro models bypass a Chat limit?**

Not necessarily. On Pro 100, GPT-6 Pro and GPT-5.6 Sol Pro currently share one weekly allowance. Switching between them does not replenish it. Other tiers have different rules. 

**Does running Codex locally make model usage free?**

No. Local messages and cloud chats share the relevant subscription allowance. Codex authenticated with an API key instead follows API pricing; using local execution does not itself remove inference charges. 

**Does an image or upload warning mean Codex is exhausted?**

Not necessarily. OpenAI currently documents separate ChatGPT upload, image-generation, and voice limits. Read the specific banner before attributing the restriction to Codex. 

**Is purchasing a reset the same as buying credits?**

No. A purchased reset immediately restores applicable allowances and changes the weekly schedule. Credits are a separate balance for eligible additional usage. A purchased instant reset cannot be banked for later. 

## Conclusion

The announced transition changes how subscribers should evaluate capacity, but its practical value requires evidence from real workloads.

**Check the current usage dashboard, record representative tasks, and compare completed work before upgrading or purchasing additional usage.** For heavy research and coding users, the strongest subscription comparison is the amount of reliable work completed within the budget—not the number of prompts sent or the presence of an unlimited label.