Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
ChatGPT and Codex will share one usage limit from December: what counts, what is exempt, and what to do
From December, ChatGPT and Codex share one usage allowance on Plus, Pro and Business. What counts, why Instant is exempt, and what to do now.
Prashant Shaw
Founder, Sumibako
7 min read
On 31 October somebody opened a thread in OpenAI's forum called "Please Keep ChatGPT and Codex Usage Limits Separate in December 2026". By 5 November it had passed 200 posts and 6,000 views, the forum's most-read thread of the week. Four matching issues had been filed against the Codex repository, and OpenAI had replied to none of them. What everyone was reacting to is three sentences in a help article.
OpenAI's usage limits article says: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience." This post is what that means in practice: what will count against the shared allowance, what will not, how much a Plus plan holds today, and what to do before December. Everything was checked against OpenAI's own pages on 3 November 2026, and the forum and the issues again on 5 November. OpenAI has not published a date in December or any new numbers, so neither do I.
🪢 What changes in December
Today, a ChatGPT subscription has two pockets. Chat has its own limits, per model and per tool, and Codex shares a second allowance with ChatGPT Work, ChatGPT's agent mode. OpenAI's Codex pricing page says it in bold: "ChatGPT Work and Codex share usage."
From December the pockets are sewn together. The help article says, in full:
One allowance, for Plus, Pro and Business, "shared across ChatGPT and Codex".
Measured in tokens. "Your usage for ChatGPT and Codex will be measured in tokens, the units of information a model processes and generates." Bigger models and longer tasks use more.
One exception. "Instant chat will not draw from your usage."
You will "be able to see your combined remaining usage and reset times in Settings → Usage."
The usual exits. At the limit, "you can wait for it to reset, upgrade your plan, or buy credits to continue using eligible features."
What the article does not say matters as much. It names Plus, Pro and Business and nothing else, so Free, Go, Enterprise and Edu are not mentioned either way. It gives no date more precise than December. And it gives no new size for the combined allowance, only a hope: "We expect most users won't need to wait for their usage to reset more often than they do now."
🧮 What will count, and what will not
Tokens are everything that passes through the model. The Codex pricing page lists them: "Your prompt, files, chat history, tool results, and ChatGPT's response all use tokens." So, once the allowance is shared:
Counts against the shared allowance
Does not
Chat with a thinking model, by OpenAI's description of tokens
Instant chat, by name
Codex in the CLI, IDE extension, desktop app and cloud
ChatGPT Work
ChatGPT for Excel on Plus and Pro, which already shares Codex's allowance today
That last row is in the pricing page already: "Usage limits are shared with other agentic features once pricing for those features is effective. This currently includes" ChatGPT for Excel. The pattern is clear. Anything agentic draws from one pool, and from December that pool includes chat with the models that think.
The line that decides your December is "chat history". In a long conversation, every new message is answered with the whole history in view, so a thinking-model chat that has run all week costs more per message than a fresh one. Until now that cost came out of chat's own pocket. From December it comes out of the same allowance as your agent.
📊 What a Plus plan holds today
OpenAI does publish estimates for Codex, which is more than it does for chat. Per five hours, on Plus and standard Business seats, for local messages:
Model
Messages per five hours
GPT-6 Astra
5 to 45
GPT-6.1 Sol
15 to 160
GPT-6.1 Terra
35 to 285
GPT-6 Luna
350 to 3,000
The page is honest about the ranges: "These estimates are not fixed message limits". Weekly limits may also apply, and "Pro plans currently have no five-hour limit." On Pro, the Pro tiers article says Pro 100 and Pro 500 "now include twice their previous Work and Codex allowance", and that on Pro 200 "The reduction announced in our September 29 email is canceled." For Plus: "Plus usage is unchanged."
So the change lands hardest on the people in that forum thread: Plus subscribers who chat with a thinking model all day and then open Codex in the evening. Whether the combined allowance will be bigger than the Codex one is today is the one number the article does not give.
😤 What people are asking for
The forum thread asks for the obvious thing, and its second post asks for the reasonable fallback: "If OpenAI decides to combine them anyway, a real-time usage dashboard must be available before the change takes effect." The most-reacted issue in the Codex repository, #58530, asks OpenAI to "preserve separate ChatGPT and Codex usage allowances on Plus", and #58723 asks it to make agent quota costs auditable. On 5 November none of the four issues had an answer from OpenAI, and the most-reacted one had 29 comments.
If you want to be heard, the thread and the issues are where OpenAI's staff read. Numbers carry further than adjectives: what you do in a typical week, on which model, and when you hit limits now.
🧯 What to do before December
1️⃣ Make Instant your default for everyday chat. It is the one thing the article exempts by name. Switch to a thinking model when the question needs it, not because it was last selected.
2️⃣ Learn where your usage goes now. The usage dashboard shows Codex today, and /status shows it inside a CLI session. A week or two of watching tells you whether you are a Plus person or a Pro person before December decides for you.
3️⃣ Pick the model per job. Astra for the hardest work, Sol for long-running work, Terra and Luna for routine and repeatable tasks. The table above is a spread of about seventy times on the same plan.
4️⃣ Trim what every Codex message carries. OpenAI's own tips: keep AGENTS.md small, and "Limit the number of MCP servers you use", because "Every MCP server adds more context to your messages and uses more of your limit." I wrote up what goes in AGENTS.md and what belongs in none of the instruction files.
5️⃣ Keep chats short. Start a new chat for a new topic and carry the useful part over in a short handoff note, not the whole history. The same habit fixes the Context window full wall.
6️⃣ Decide between credits and an upgrade in advance. Plus and Pro can buy credits at the limit; Pro 100 is $100 with a doubled allowance. Either is cheaper chosen calmly in November than at 11pm in December with a task half done.
⚖️ Claude has always worked this way
None of this is unusual. On Claude, chat and Claude Code have shared one pool from the start: "Your activity across Claude on web, desktop, mobile, and Claude Code all draws from the same pool", per its pricing page. What OpenAI is changing is not the model of the world; it is the size of the pocket people got used to. If you are weighing the two, I compared them this week in Claude Code vs Codex, and Claude's own limits are in its own post.
📍 The cheapest token is one you do not resend
Most of the advice above is the same advice: send the model less history. The state of a piece of work belongs in a document both your chat and your agent can read, not in a conversation that grows until it is expensive.
That is what I built Sumibako for, so weigh this paragraph accordingly. Codex files a page into it with one command, npx sumibako, on the free plan, and reads it back in the next session. On ChatGPT, a custom connector can write to it on Business, Enterprise and Edu, and only read on Pro; on Plus you paste the page in, which costs a few hundred tokens instead of a week of history.
OpenAI is merging two meters into one and promising the bill will feel the same. Every household that ever combined two phone plans has heard that sentence, and some of them were right.
