OpenAI Doubled Work and Codex Limits on Every Pro Tier, and Killed a Cut It Had Already Announced

OpenAI has doubled the Work and Codex usage allowances attached to every paid Pro tier and to Business Premium, effective immediately and at unchanged subscription prices. The company also reversed a reduction it had already communicated to customers, which would have cut the Pro 200 allowance roughly in half. The change was announced on October 29 by Tibo Sottiaux, who leads Codex and ChatGPT Work, and is now reflected in OpenAI's help center documentation.

The mechanics are more uneven than the word "doubling" suggests. Pro 100 at 100 dollars a month and Pro 500 at 500 dollars a month both get twice their previous Work and Codex quota. Pro 200 is the complicated case: OpenAI had planned to cut that tier's shared Work and Codex allowance from 20 times the Plus allowance down to 10 times, and to halve the weekly GPT-6 Pro message cap in regular chat from 200 to 100. Both cuts are cancelled. Subscribers who already had the higher allowance keep it, and subscribers on the lower allowance get doubled up to the same level. Two people paying 200 dollars can therefore both be right when one says nothing changed and the other says the quota doubled.

What did not move is just as important. The weekly GPT-6 Pro message limit in ordinary chat is unchanged, and ChatGPT Plus limits are untouched. The extended allowances for Dots, OpenAI's always-on background agents, remain in place, with Dot usage still metered separately from the default Work and Codex pool. Existing subscribers receive the new limits automatically without re-subscribing, and OpenAI has not attached an end date to the increase.

The transparency gap is real. OpenAI has not published absolute quota numbers for any tier, so customers cannot verify a doubling against a documented figure, and the change lands as a directional promise rather than a measurable contract. The company also cautions that doubling a quota does not mean twice as many completed tasks, since consumption varies by model, task type and settings.

The timing explains the generosity. Anthropic has filed to go public on Nasdaq and is competing for the same enterprise buyers, and the two labs are now fighting over the same pool of institutional capital as well as the same developers. OpenAI has also been shipping cheaper, more efficient models, including the GPT-6.1 Terra and Sol releases, which raise usable capacity without adding hardware. Doubling quotas is a way to convert that efficiency into goodwill at the exact moment a rival is asking investors to compare the two businesses.

It is also a small admission. The original plan to cut Pro 200 was a straightforward margin move, and the reversal shows how quickly capacity has become a competitive lever rather than an internal cost line. A pricing page that once crept upward now moves in the other direction, in public, within days of a backlash.

The lesson for buyers is to stop treating published limits as fixed inputs to procurement planning. Quotas on the frontier tiers are now a marketing variable that can move twice in a month, in either direction, and the only durable defense is workload portability: keeping prompts, evaluation harnesses and agent scaffolding portable enough that a capacity surprise does not become an architecture decision.