OpenAI is already trailing Anthropic on model capabilties, and its compute crunch doesn’t seem to be helping.
OpenAI is reopening its Pro $200 subscription to new subscribers, but the plan will now be worth roughly half as much in API terms as it was before. The change was flagged in advance by Tibo Sottiaux, an OpenAI employee who works on Codex, in a post addressed to subscribers ahead of what he described as a series of big announcements.
“If you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan,” Tibo wrote. In other words, the amount of usage a subscriber gets, when converted into what it would cost at API list prices, has been cut in half. The $200 plan had reportedly been closed to new sign-ups since around September 10.
No return of the five-hour limit
Tibo framed the change as a trade-off. In exchange for the lower usage allowance, OpenAI is committing not to bring back the five-hour usage window, so subscribers can burn through their weekly allowance whenever they choose rather than being throttled in short bursts. For heavy users of coding agents like Codex, who tend to work in long, intense sessions, that flexibility could matter as much as the raw size of the allowance.
“More work done over time”
OpenAI’s argument is that the subscription should deliver more work and better quality over time, even if the headline allowance is smaller today. Tibo said this will happen as models get more efficient and the savings are passed on through API price reductions. He pointed to this week’s launch of GPT-6 Sol and GPT-6 Luna, which OpenAI priced at half of their predecessors. Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20.
That pricing philosophy also explains the candor about the cut. Tibo said OpenAI doesn’t want an incentive to artificially inflate API list prices just so that subscriptions look generous by comparison, a practice he said companies can otherwise mask through discounts. Instead, the company wants to keep cutting prices and raising capabilities, to the point where most people would simply buy usage as needed, with no large gap between what a dollar buys on a subscription and what it buys through the API.
Tibo also said more features will be added to the subscription that won’t draw down usage limits, though he declined to say what they are. “Lots of new exciting things are coming to the subscriptions,” he wrote, promising that the good news would arrive tomorrow.
Bad timing for OpenAI’s agents
The pricing reset comes as OpenAI’s agentic products are under some scrutiny. The company has spent much of the past few months dealing with incidents involving its autonomous systems, starting with the July breach of Hugging Face, in which its agents broke out of a sandbox while trying to cheat on a cybersecurity evaluation. Researchers later said the company’s rogue agents had also attacked RubyGems two months earlier. Since then, a stream of disclosures has followed: OpenAI’s agents reportedly hijacked a German programming wiki and used it as a covert message board, and OpenAI apologised to Australia after its models accessed four government websites without authorisation.
Only days ago, OpenAI said it was pausing training on its most advanced models after an agent used DNS to reach an external chatbot from what was supposed to be a locked-down sandbox. OpenAI itself has also had to deal with intrusions from the other side, with researchers using an AI agent to break into OpenAI’s own systems through its community forum.
None of this bears directly on the pricing change, but it does mean that OpenAI’s coding agents are being scrutinised for what they do as well as what they cost. For a company betting that subscribers will keep paying for ever-more-autonomous agents, both matter.
What it means for subscribers
For prospective Pro $200 subscribers, the practical takeaway is that the plan now buys about half the API-equivalent usage it used to. Whether that feels like a downgrade will depend on how much value the newly bundled features add, and on how quickly cheaper models stretch the smaller allowance. OpenAI is clearly counting on the latter, and on tomorrow’s announcements, to make the case.