OpenAI reopens $200 Pro plan while halving API credits

By Splitfeed EditorialLast updated Sep 29, 2026, 10:20 UTC

On September 29, 2026, OpenAI reopened its $200-per-month Pro subscription to new sign-ups after a period of restricted access. At the same time, the company cut in half the API credits included per dollar in the plan. OpenAI employee Thibault Sottiaux offered the company's rationale publicly, pointing to newer, more efficient models, specifically GPT-6 Sol and Luna, as compensation for the reduced credit value.

The credit cut is a concrete change in what subscribers receive for their money, and it arrives alongside a stated strategic pivot away from subsidized flat-rate pricing toward usage-based billing. For users who depended on the Pro plan primarily as a cost-effective route to API access, the halving of credits represents a real reduction in value regardless of model efficiency gains. Whether efficiency improvements in GPT-6 Sol and Luna actually offset the difference in practice depends heavily on what tasks a given user runs and how those models perform on them, which makes the tradeoff genuinely difficult to evaluate in the abstract.

None of the three camps have published reactions to this story yet. Pro-AI commentators would typically frame the credit reduction as a reasonable business adjustment, pointing to Sottiaux's efficiency argument and treating the plan's reopening as a net positive for access. Anti-AI voices would likely focus on the credit cut as evidence that OpenAI is quietly degrading subscriber value while maintaining the same headline price, part of a broader pattern of prioritizing revenue over user benefit. Middle-ground observers would probably acknowledge both sides, noting that efficiency claims are plausible but unverified, and that a shift toward pay-per-use billing favors heavy power users while squeezing those who budgeted around flat-rate predictability.

The argument will sharpen once developers and researchers publish benchmarks comparing GPT-6 Sol and Luna's output per effective credit against previous models under the old pricing structure. Independent cost-per-task analyses, rather than OpenAI's own framing, will be the most useful signal for whether Sottiaux's efficiency claim holds up in real workloads.