OpenAI Reopens $200 Pro Plan, Halves API Credits in Shift to Pay-Per-Use

OpenAI has reopened its $200-per-month Pro subscription to new sign-ups while cutting the API credits attached to each dollar of that plan in half. The company says the reduction does not leave subscribers worse off, because its newer, more efficient models — GPT-6 Sol and Luna — deliver enough of a performance-per-dollar improvement to compensate for the smaller credit balance.

The announcement came from Thibault Sottiaux, an OpenAI employee, who presented the efficiency argument as the justification for the change. Users who sign up for the Pro plan again will therefore get half the API credits they would have received under the previous structure, but OpenAI's position is that each credit now buys substantially more model capability.

What changed

Two things happened at once. First, the $200 Pro plan, which had been closed to new sign-ups, is available again. Second, the API credits bundled with that plan have been reduced by 50 percent. OpenAI frames the pair of moves as a rebalancing rather than a price increase: same subscription fee, fewer nominal credits, more capable models per credit.

The company's stated reasoning, delivered by Sottiaux, is straightforward. Models like GPT-6 Sol and Luna are efficient enough that the effective compute a subscriber can extract from the plan has not declined, even though the headline credit number has halved.

The shift away from flat-rate subsidies

The credit cut is part of a broader pattern in how OpenAI prices access to its models. The company is moving away from subsidized flat-rate plans and toward usage-based billing — a pay-per-use structure in which what a customer pays tracks what a customer consumes.

That transition carries real weight for the AI market. Flat-rate subscriptions shift financial risk onto the provider: heavy users can extract far more value than their subscription fee covers, and OpenAI absorbs the difference. Usage-based billing reverses that arrangement, tying revenue directly to consumption and making unit economics more predictable for the provider.

For developers and power users, the implications are equally concrete. Under a pay-per-use model, costs scale with workload. A team running high-volume inference pays more than a team running occasional queries. The subsidized flat-rate era, in which a fixed $200 could buy outsized access, is what OpenAI is now winding down.

Why efficiency is doing the heavy lifting

OpenAI's argument rests on a specific claim: that GPT-6 Sol and Luna are efficient enough to make half the credits go as far as the full allocation did before. This is the mechanism that lets the company cut credits without calling it a price hike.

If the efficiency gains hold up in practice, subscribers get equivalent or better output for the same $200. If they do not — if real-world workloads consume more tokens or more compute than the efficiency math assumes — then the halved credit pool becomes an effective price increase for those users. OpenAI's framing, via Sottiaux, is that the gains are real and the trade is neutral for the subscriber.

The reopening of the Pro plan itself is notable. Closing a paid subscription tier to new customers is unusual, and its return suggests OpenAI now has the capacity — or the pricing structure — to support new sign-ups again. The plan comes back on different terms than the one that left, with the credit structure reset to reflect the new billing philosophy.

What it means for the market

The stakes here extend beyond one subscription tier. OpenAI's pricing decisions function as a signal to the entire AI industry, from competing labs to the developers choosing where to build. A decisive move toward usage-based billing from the category's dominant consumer and developer platform suggests the industry-wide era of loss-leading flat-rate access is receding.

For OpenAI, usage-based billing aligns revenue with the cost of delivering inference. As models grow more capable and more expensive to serve, that alignment matters more. Subsidized plans that made sense as customer-acquisition tools become harder to sustain at scale.

For subscribers, the message is equally clear: the value proposition of a fixed-price plan now depends on model efficiency improvements arriving fast enough to offset reductions in nominal credits. GPT-6 Sol and Luna are the current justification. Whether future efficiency gains continue to pace future credit adjustments will determine whether the $200 Pro plan stays a good deal or quietly becomes one.

OpenAI has not framed the change as a reduction in value, and Sottiaux's efficiency argument is the company's official line on why it isn't. The next data point will come from subscribers themselves, as they measure what half the credits actually buy against GPT-6 Sol and Luna in production workloads.