ChatGPT chats will soon eat into Codex coding time

OpenAI is changing how its paid ChatGPT plans count usage. Its Help Center article on usage limits, updated October 31, says that "starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex," as CellCog quotes it. Codex is OpenAI's coding agent: a program that reads, writes and tests code on its own. Right now, a long chat and a long coding job mostly draw on separate limits. After the change, they drink from one glass.

The article says that pot will be counted in tokens, "the units of information a model processes and generates." Quick replies in Instant chat, ChatGPT's fastest mode, will not count. When the pot runs dry, the choices are to wait for it to reset, upgrade the plan or buy credits. AI TL;DR's summary notes that OpenAI has not named a day in December or said how big each plan's allowance will be.

Part of this is already true. OpenAI's pricing page says ChatGPT Work, the agent features inside ChatGPT, already shares usage with Codex. The same page says making a picture "counts toward the same general usage limits" and uses them up about three to five times faster than a turn without one. December widens the shared pot to ordinary chat.

That lands right on Jon's system. His fleet hands a lot of its coding to Codex workers on his ChatGPT plan, run by Ringer, the orchestrator he built to launch and check them. The picture at the top of this story, like every News Factory cover, came from Codex's built-in image tool on that same plan. From December, a long planning chat on that account will come out of the same allowance as those workers and those covers.

Some users are already pushing back. A GitHub issue opened November 1 asks OpenAI not to merge the limits, arguing that planning in chat "drains the exact quota needed to implement the code in Codex." An earlier issue asks to keep the Plus allowances separate. Neither reports a measured loss, since the change has not started. OpenAI says it expects "most users won't need to wait for their usage to reset more often than they do now." That is a forecast, not a measurement.

How to use it

• Once the change lands, the combined balance and reset times show up in ChatGPT under Settings > Usage, according to DEV Community's write-up.
• Inside a running Codex command-line session, typing /status shows the limits left, per the pricing page.
• Pictures are the expensive part. At three to five times the cost of a normal turn, a batch of images can eat a day's allowance fast.
• Work run with an API key, a paid developer key, is billed separately. The pricing page says "API token prices are separate from subscription usage."
• The catch: the pricing page still says Pro "currently" has no five-hour limit. Nothing says whether that survives December.

What nobody knows yet

The exact start day. How many tokens each plan gets. Whether Pro keeps its open five-hour window. Whether a picture made in chat costs the same as one made in Codex. Whether OpenAI's forecast about waiting holds up for people who chat and code heavily on one account.

What to watch

OpenAI's Help Center article and the Settings > Usage screen as December starts, the pricing page for new per-plan numbers, and any staff reply on issue 58655.