OpenAI has made its heavier ChatGPT plans bigger and cancelled a cut that would have started on October 30. The Work and Codex allowance on Pro 100 and Pro 500 is now twice what it was, Pro 200 accounts that had the smaller allowance were raised to the larger one, and Plus is unchanged. Codex is OpenAI's coding assistant.
The cut was announced in September, and we covered it then. What follows from the reversal is the more useful part.
The cut would have erased the reason to pay $200
Pro 200 is described as a multiple of the $20 Plus plan. Before the cut it was 20 times Plus, and the cut would have made it 10 times Plus. Do the arithmetic: at 20 times, each Plus-sized helping of work costs $10. At 10 times, it costs $20, the same as Plus. Ten times the work for ten times the price is no discount at all, so the cut would have taken away the whole volume discount that made Pro 200 worth buying.
OpenAI's case in September was that its models are getting more efficient, so a smaller multiple would still get the same work done. That may well be true over a year, but a subscriber feels the allowance this month, and the people who felt it loudest were the ones who use the tools most.
A bigger allowance is not twice the tasks
OpenAI itself says that doubling an allowance does not double the finished work, because a task's use of the allowance depends on the model, the kind of task and the settings. A long coding job on the biggest model can use many times what a short question on a small one uses. So the allowance is a pool of capacity, and the number of tasks it covers depends on how efficient the model is that day. That is why the newer, more efficient models matter more to a heavy user than any headline multiple.
The same goes for the always-on Dot agents OpenAI launched in September. Their use is counted separately for now, but work a Dot hands to Codex still draws from the main pool.
Who pushed, and why it worked
AT&T's customers could not easily leave, because Verizon was cutting too. Subscribers to AI tools are in a different position. Many developers use Codex and Anthropic's Claude Code side by side and compared the two openly when the cut was announced. When switching costs an afternoon, the heaviest users are the ones a company can least afford to annoy. I expect the labs to keep competing on allowances in this way: raising them when a rival gains ground, and holding them when it does not.
Count finished jobs, not plan names
Take a five-person web agency that pays for one Pro plan so that one developer can use Codex to fix client sites. Suppose the plan covers about 80 finished fixes a month, and a change of limits cuts that to 40. The price is the same $200, but the cost of one fix has gone from $2.50 to $5. Nothing on the invoice shows it.
So the useful habit is to keep a simple count: how many jobs your team finished with the tool this month, and what each cost. Then a change like this one shows up as a number you can act on: stay, buy a bigger plan, move steady work to pay-per-use, or try the rival. The price will stay put, but the work it buys will not, so measure the work.