ChatGPT and Codex Shared Usage Limit Explained (2026)
From December 2026, ChatGPT and Codex share one usage allowance on Plus, Pro and Business plans. Usage is counted in tokens. Instant chat does not draw from it; thinking models, Work tasks and Codex sessions do. OpenAI has not named the start day or the allowance size per plan. Here is what is confirmed, what is reported, and what to do before December.
Bottom Line
From December 2026, one allowance covers ChatGPT and Codex on Plus, Pro and Business. It is counted in tokens. Instant chat is outside it. Everything agentic is inside it.
| Question | Answer (as of 6 November 2026) |
|---|---|
| Which plans? | Plus, Pro and Business |
| When? | December 2026. No day is named |
| Measured in? | Tokens, not messages |
| Instant chat | Does not draw from the allowance |
| Thinking models | Draw from it |
| Work tasks | Draw from it (Work already shared Codex’s pool) |
| Codex (CLI, IDE, cloud, desktop app) | Draws from it |
| Where to check | ChatGPT Settings → Usage: combined remaining usage and reset times |
| At the limit | Wait for the reset, upgrade, or buy credits for eligible features |
| Allowance size per plan | Not published |
| Free and Go | Not named in the reports of OpenAI’s article |
OpenAI’s own line, as quoted by Notebookcheck: most users “won’t need to wait for their usage to reset more often than they do now.”
What actually changes
Before this change there were two pools on a paid plan:
- The chat pool. Messages to ChatGPT models, with their own caps.
- The agent pool. OpenAI already stated that “Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents use a shared allowance and credit pool.” We covered that in Claude Code vs Codex limits compared.
From December the chat side joins the agent pool, with one exception: Instant chat stays free of the meter. A thinking-model chat now spends the same tokens your Codex session needs.
The reason OpenAI gives is product, not price. The help article ties the change to bringing “the agentic capabilities of Work into the main ChatGPT experience.” SQ Magazine reported a subscriber email on 2 November 2026 that announced the Work tab will fold into regular chat, starting with select users. Codex stays a separate experience in the ChatGPT desktop app.
Who should care
| You use… | Effect from December |
|---|---|
| Codex only, chat rarely | Little change. Your Codex sessions were already the main draw |
| Instant chat plus Codex | Little change. Instant chat does not count |
| Thinking-model chat and Codex every day | Real change. Long thinking chats now eat the window your agent needs |
| Work tab for long tasks | Same pool as before. Work already shared Codex’s allowance |
| Codex through an API key | Not affected, per OpenAI’s Codex pricing page as reported by Pillitteri. API usage bills separately |
The line most coverage misses: for heavy Codex users, the agent pool was already shared. The new cost is the thinking-model chat you run beside the agent. If you research in a thinking chat and then hand the plan to Codex, both halves now come out of one balance. Move the research to Instant chat where the quality is enough, and save the tokens for the agent run.
The five-hour and weekly windows
OpenAI has not said how the five-hour and weekly windows apply to the merged pool. What two secondary sources (Pillitteri and Datastudios) report about the rules today:
- Plus and Business: a five-hour window and weekly limits.
- Pro: no five-hour limit at the moment. Only the weekly cap counts.
- Enterprise and Edu: by agreement, no fixed rate limits (Datastudios only).
The same two sources list current Codex local-message ranges per five hours on Plus. They are the clearest view of how differently each model spends the pool:
| Model | Local messages per 5 hours, Plus (reported) |
|---|---|
| GPT-6 Astra | 5-45 |
| GPT-6.1 Sol | 15-160 |
| GPT-6.1 Terra | 35-285 |
| GPT-6 Luna | 350-3,000 |
These ranges are second-hand. OpenAI’s help article returned HTTP 403 to us on 6 November 2026, so we could not read them at the source. The spread is the point: the largest model can spend the pool about 70 times faster than the smallest (our ratio of the range ends: 5 vs 350, 45 vs 3,000). That ratio carries into a token-based shared allowance.
How to find your own number
Nobody outside OpenAI knows the December allowance. You can measure your own burn rate before it lands:
- Open ChatGPT Settings → Usage at the start of a work block. Note the remaining percentage and the reset time.
- Do a normal session: your usual mix of chat and Codex.
- Read the panel again. The change per hour is your burn rate.
- Repeat with Instant chat for research instead of a thinking model. The difference is what the merge would cost you.
- In a Codex terminal session, the
/statuscommand shows usage, per Pillitteri. We did not confirm this against OpenAI’s docs.
Your numbers will not transfer to anyone else’s account. The token cost depends on the model, the context length, and how long each agent run goes.
What is still unknown
- The start day in December.
- The token allowance for each plan.
- Whether the five-hour window changes for Plus and Business when chat joins the pool.
- Free and Go. The reports of OpenAI’s article do not mention them. Pillitteri reports that Free and Go include Codex with GPT-6 Luna in the desktop app. That claim has one source.
Treat any post that quotes a December token number as a guess until OpenAI publishes one.
If you want a coding agent with no allowance
A coding agent that runs on a local model draws from no vendor allowance at all. The trade is hardware cost and model quality. Can Qwen3.8 replace Claude for coding? covers how close a local model gets for agent work, and the gear list covers the machines by memory tier.
Sources
- OpenAI Help Center — Understanding ChatGPT and Codex usage limits (returned HTTP 403 on 6 November 2026; quotes above come from the reports below)
- Notebookcheck — ChatGPT and Codex will share one usage limit starting in December (3 November 2026)
- Pasquale Pillitteri — OpenAI confirms one shared usage limit for ChatGPT and Codex from December (1 November 2026)
- Datastudios — ChatGPT and Codex shared usage limit from December: Plus, Pro, Business (early November 2026; no absolute date shown)
- SQ Magazine — ChatGPT and Codex to share usage limits as Work tab merges (2 November 2026)
- Hacker News — ChatGPT and Codex usage limits merging (discussion, 31 October 2026)
See Also
- Codex Free Reset and Banked Reset (2026) — every recent reset by date, and how a banked reset works
- Stop Hitting Codex Usage Limits —
config.tomlsettings and token habits that stretch the window - Claude Code vs Codex Limits Compared — how the two vendors pool usage across surfaces
- Is Codex Fast Mode Worth It? — what fast mode costs you against the window
- Subscription Agent Harness Rules for Claude and ChatGPT — what plan terms allow when agents run on a subscription