ChatGPT and Codex Will Share Usage Limits in December 2026: What Changes

Verified 2 November 2026 | Announcement surfaced 1 November 2026 | Effective: December 2026, exact day not yet specified | Status: announced, not confirmed live.

If you use ChatGPT to research, Codex to build software, and Work to complete longer projects, a subscription change is coming that deserves more attention than another model launch. OpenAI says that ChatGPT Plus, Pro and Business will share one usage allowance across ChatGPT and Codex starting in December. Instant chat will not draw from that allowance. The announcement appears in OpenAI's official usage-limits guidance and was brought to its developer community on 1 November 2026.

What OpenAI has actually confirmed

OpenAI describes a combined allowance for Plus, Pro and Business subscribers, measured in tokens. Usage depends on the model selected and the complexity of the task; more advanced models and demanding tasks generally consume more. The company says that Instant chat is excluded from this shared allowance. Subscribers will be able to inspect their combined remaining usage and reset times under Settings > Usage. At a limit, the documented options are to wait for reset, upgrade, or buy credits for eligible features. These are the confirmed rules in the primary OpenAI help article.

The change accompanies OpenAI's effort to bring Work's agentic capabilities into the main ChatGPT experience. Today, Work handles multi-step research and deliverables, while Codex focuses on software engineering. OpenAI's Work and Codex documentation describes those different tasks and the existing usage structure. The December announcement concerns metering across experiences; it does not say that Codex and Work become the same product or that every task will consume an identical number of tokens.

What is not yet known

OpenAI has not specified the exact December switch-over day, the final included token allowance for each named plan, or a complete before-and-after comparison of heavy-user capacity in the announcement. Its statement that most users are not expected to wait for resets more often is an expectation, not a guaranteed outcome for every subscriber. The official usage guidance does not provide enough information to calculate whether a particular developer or analyst will receive more or less effective capacity.

Do not convert the announcement into an unsupported claim that all paid subscriptions are being cut, that every ChatGPT message will be charged, or that Enterprise and Edu are covered by this exact Plus/Pro/Business notice. Equally, do not promise that current usage patterns will remain unaffected. Those conclusions require the final plan allowances and measured workloads, which have not been published in this announcement.

Why the change matters to teams using AI agents

Consider a consultant who uses a high-effort ChatGPT conversation to design an integration, Work to assemble a test report, and Codex to implement the connector. These are different business tasks, but a shared usage budget can make their combined demand matter more than any single tool's activity. The practical planning question is no longer just 'How much Codex do I use?' It is 'How much capacity do all my advanced AI workflows consume together?' OpenAI already explains that task length, model selection and reasoning effort affect usage in its product guidance.

For SAP and HR technology teams, this is especially relevant when the same subscription supports solution design, API debugging, data-mapping scripts and release-test documentation. A successful proof of concept can become a recurring workload. Budgeting for that workload should account for the total work completed, not just a count of chat messages. This is an architectural planning implication, not a claim that OpenAI has measured usage in any SAP SuccessFactors environment.

Five checks to make before December

1. Map who uses which plan. Identify Plus, Pro and Business users, and distinguish personal subscriptions from company workspaces. Do not assume an Enterprise contract follows the same announcement.
2. Record a representative workload baseline. For a normal week, note which tasks use advanced ChatGPT, Work and Codex, which models they select, and where long-running or repeated work occurs. This is an internal measurement exercise, not an OpenAI-provided token forecast.
3. Identify work that can use Instant. Simple drafting, explanations and low-risk questions may not need the most demanding model. OpenAI explicitly exempts Instant chat from the forthcoming shared usage allowance; check that the experience you select is actually Instant.
4. Separate subscription credits from API credits. OpenAI's promotional API-credit benefit is a different program and does not increase signed-in ChatGPT or Codex allowance. Do not count an API grant as additional capacity for your ChatGPT plan.
5. Recheck the live dashboard when the transition starts. Use Settings > Usage for the displayed remaining allowance and reset time. If an agent is essential to a business process, define a fallback or credit budget before a limit interrupts delivery. See OpenAI's personal-plan credit guidance for supported credit use, subject to plan eligibility.

A simple acceptance test for AI-heavy workflows

I would run the same small set of work across two comparable weeks: a document-analysis task, a repository change with tests, and a longer multi-step research or implementation task. Capture task completion, elapsed time, selected model, retries, and any displayed allowance change. Repeat after the December transition using the same task definitions and account tier. If the new usage display changes how work is metered, document that rather than assuming old message-based comparisons still apply. The purpose is to measure usable capacity per completed outcome, not to manufacture a dramatic percentage.

Business administrators should also decide which work must finish even if a user reaches a limit. A critical deployment, payroll integration or release validation should not depend on an unmeasured personal AI subscription. Keep code review, access controls and production approval outside the assumption that an AI tool will always be available. OpenAI's announcement establishes the upcoming metering change, not service-level guarantees for those workflows.

The bottom line

The verified news is one shared, token-measured allowance for ChatGPT and Codex on Plus, Pro and Business starting in December 2026, with Instant chat exempt. The exact effective date and final per-plan allowances remain unspecified in the announcement. The sensible response is to establish your current workload baseline, keep API credits separate, and revisit actual capacity once the new Settings > Usage figures appear. OpenAI's 1 November community notice and official help page are the sources to watch for revisions.