Starting in December, ChatGPT and Codex Will Share a Quota: Instant Chat Doesn't Count, Coding Has to Compete

OpenAI Help Center updated a rule: starting in December, ChatGPT and Codex for Plus, Pro, and Business will share a single token quota. Instant chat does not count against the quota. The specific numbers were not provided. The link points to the OpenAI Help Center. The original message came from an X post by user JkwuRmsl.

The attached image is a screenshot of a report. The headline reads 'OpenAI Will Combine ChatGPT and Codex Usage Allowances in December,' and the subheading mentions that Plus, Pro, and Business share a token-based quota, with Instant chat not counted. The author is Simon Yoon, and the publication date is October 31, 2026.

For ordinary users, the Instant chat exception is key. Everyday short Q&A and quick search-style conversations do not enter the quota pool, so the entry-level experience remains unchanged for now. But whether long conversations, files, and complex tasks are counted is not detailed in the material. For developers, Codex agent tasks consume tokens; coding and running tasks share the same pool as chatting, so competing for the same allocation is more direct. Previously, you worried about agents running wild; now you also have to worry about chat eating up the pool. And vice versa.

The specific quota has not been announced, so any judgment about whether it is 'enough' is premature. With one sentence in the Help Center, users have already started doing the math. The most troublesome part is not how large the quota is, but that it is impossible to estimate: for a long conversation or a batch task, exactly how much will be deducted? Without numbers, you can only wait for actual usage. OpenAI excluding Instant chat shows that light chat is the first thing to protect. What is truly expensive is long context and agents. Are you more afraid of chat using it all up, or agents using it all up?