OpenAI reopens its $200 Pro tier but halves the API credits per dollar
OpenAI is selling its heaviest consumer plan again, on cheaper terms for the company and tighter ones for the buyer — and it has finally said out loud where it thinks subscription pricing is heading.
OpenAI is reopening new sign-ups for its $200-a-month ChatGPT Pro tier from September 30, but the plan now bundles roughly half the API spend it used to, and the five-hour usage cap is not coming back. Tibo Sottiaux, the OpenAI engineer who announced the original pause on X earlier this month, framed the trade plainly: the change "will net out at half the dollar in API spend compared to the old Pro $200 plan," he wrote, while subscribers "will still get more work done" than on the same plan a month ago. His evidence is OpenAI's own price cuts — GPT-6 Sol and GPT-6 Luna shipped this week at 50 percent of their predecessors' API price, so a dollar now buys more tokens than it did in August.
The mechanics matter more than the sticker. Pro at $200 carries 20x the usage of Plus; the $100 Pro tier carries 5x. Removing the five-hour ceiling means a subscriber can burn the whole weekly allowance in one afternoon of agent runs instead of being throttled, which is precisely the workload OpenAI was losing money on when it closed the tier on September 10. We covered that closure as a capacity story — OpenAI is turning away new $200 Pro subscribers as Astra demand soars — and the reopen, with half the API credits attached, is the company's answer: keep the customers, stop subsidising their inference bill.
The strategic line is the one Sottiaux wrote out in full. OpenAI wants API prices to fall far enough "that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent." That is a description of a world where the subscription stops being a discount and becomes a convenience wrapper — metered billing with a familiar logo on it. Microsoft has already moved Copilot in the same direction, and it is the obvious end state for labs whose compute costs scale with use. Read the reopen as the first visible step from flat-rate AI to pay-per-use AI, taken by the company with the most to gain from it.
Freedom of information documents show a six-month live facial recognition trial across London railway stations scanned more than half a million faces, cost taxpayers £320,786 and produced exactly one alert — which was wrong. The British Transport Police ran 18 deployments between February and July at some of the capital's busiest stations, according to the FOI response obtained by Liberty Investigates and shared with The Guardian, and logged no arrests from a recognition hit. The cost and police-hours figures are single-sourced to that FOI request; the operational outcome — over 530,000 faces, zero true matches, one false identification, no arrests — was separately reported by the BBC in August from BTP's own deployment logs. British Transport Police told the Guardian it made "a number of associated arrests" during the deployments for assault, theft, weapon possession and breach of court orders, but says those did not follow from an LFR alert and are excluded from the technology's performance data.
The force is expanding anyway. BTP extended the trial by four months and moved it onto London Underground stations in August, and says the extension has since produced three confirmed alerts against people complying with sexual harm prevention orders or other court conditions. Fraser Sampson, a former UK biometrics and surveillance camera commissioner, called the original six months "not very fruitful" and noted the difference in what success looks like: in a shop, deterrence is the win, but for police trying to catch people, "success means people being caught." Labour MP Bell Ribeiro-Addy said the rollout should be suspended until stronger safeguards and a legal framework exist. We covered the Underground expansion when it was announced — UK police bring live facial recognition to the London Underground — and this is the first hard results sheet for the programme that expansion is built on.
That is the uncomfortable arithmetic of these trials. A false-positive rate of one in 500,000 looks excellent on a slide and is nearly meaningless as a justification, because the number that decides whether a surveillance deployment earned its budget is how many people it actually caught, and for this trial that number is zero — against £320,786 and roughly 100 police hours that could have gone somewhere else. Liberty Investigates says more than half of police forces in England and Wales have now deployed the technology, which means a lot of forces are scaling a capability whose only completed trial found no one.
What to watch: whether BTP publishes the Underground extension's numbers on the same terms, and whether the three confirmed alerts it has already cited are enough to justify the wider rollout.
If a police trial this expensive produced no correct matches, what should it take to justify deploying the technology to a million people a day? Tell us in the comments.
Sources: X/@thsottiaux · OpenAI Help Center · Unite.AI · Android Authority · The Decoder · The Guardian · BBC News
