OpenAI is reopening its $200-a-month ChatGPT Pro plan to new subscribers tomorrow, after almost three weeks closed. The catch is the usage allowance. Thibault Sottiaux, who leads Codex and ChatGPT work at OpenAI, posted on X today that the company is "changing how we calculate the usage" and that, "if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan."
So the price is the same, but the allowance measured in API-equivalent spend is roughly halved. If your team relies on Pro seats for Codex or heavy ChatGPT use, this matters.
Why the plan was closed
On 10 September 2026, OpenAI stopped new sign-ups and upgrades to Pro $200, CIO reported. Existing subscribers kept their plans, but anyone who cancelled or downgraded could not come back while the pause lasted. The cheaper Pro $100 tier, which OpenAI introduced in April with five times the usage of Plus, stayed open.
The stated reason was demand for GPT-6 Astra, released on 4 September. "Demand for Astra is really unprecedented," Sottiaux told CIO. "We're pulling all the levers possible to sustain the demand." Fortune reported that the $200 plan put the most strain on OpenAI's systems, and that Astra's computer-use features used up allowances faster than earlier models. Sottiaux described the pause as "the smallest step that allows us to continue giving the broadest access possible." TokenPost noted at the time that OpenAI had not officially confirmed the cause, so the explanation rests on Sottiaux's comments.
In plain terms, OpenAI ran short of compute and limited the plan whose users were most likely to max out their allowances.
What OpenAI is offering in return
Sottiaux argues that subscribers will still "get more work done than if you were on the Pro $200 subscription one month ago." The post gives four reasons.
No 5-hour limit. In August, OpenAI brought back the rolling 5-hour cap on Codex and ChatGPT Work for Plus users, but kept it off for both Pro tiers "for the upcoming months". Sottiaux now says OpenAI is "committing to not reintroducing the 5h limit" on Pro $200, so the whole weekly allowance can be used whenever you like.
Cheaper models. GPT-6 Sol and GPT-6 Luna launched on 22 September at half the price of the 5.6 models, according to TechCrunch. Sol costs $2 per million input tokens and $10 per million output tokens, and Luna costs $0.10 and $0.50. Because the Pro allowance is now counted in API-price terms, cheaper models mean each dollar of allowance goes further.
No inflated list prices. OpenAI says it does not want to prop up API prices to make subscriptions look generous. Sottiaux expects prices to fall low enough that "it makes sense for most to buy usage as needed," with little difference between subscription and API value per dollar.
Extras that don't use the allowance. Something new is being added tomorrow that won't count against usage. Sottiaux did not say what it is.
How it has gone down
Early reactions on Hacker News are mostly sceptical. Several commenters see it as a price rise in all but name. One asked whether "500$ is the new 200$", and another said OpenAI has so far only raised prices and reduced usage quotas. A few say they were already hitting the weekly limit on the old plan. Some mention Anthropic's Claude models as the alternative they are considering.
What this tells founders about AI capacity
Flat-rate AI plans can change. In one quarter, OpenAI has reinstated time-window limits for Plus, paused sign-ups for its top tier, and changed how Pro usage is counted. None of this was hidden, but none of it was predictable either. A $200 seat is a price for access, not a promise of a fixed amount of work.
Compute is still the bottleneck. When a strong model launches, the heaviest users feel the squeeze first. If your product relies on OpenAI capacity, remember that the top consumer tier was rationed when demand spiked. The analyst advice CIO carried for businesses was to get committed throughput in writing and keep backup models for critical workloads.
OpenAI is steering people towards the API. It is openly saying that pay-as-you-go API pricing and subscriptions should end up at similar value. That makes costs easier to compare, and it makes a flat-rate subscription less of a bargain.
What to do this week
Audit your Pro seats. Check which team members actually use most of their weekly allowance. Anyone who doesn't may be fine on Pro $100 or Plus, depending on their usage.
Compare against the API. For developers who mainly use Codex, estimate their monthly token usage and price it against Sol's API rates. With the allowance now expressed in API-dollar terms, this comparison is straightforward.
Wait for tomorrow's details. Sottiaux promised "big announcements tomorrow". Don't buy or cancel annual seats until you know what the non-metered extras are.
Don't rely on a single vendor. Make sure your coding workflow can switch between OpenAI's Codex and a second option, such as Anthropic's Claude Code or an editor like Cursor that supports several models. Then a pause or quota change becomes a small annoyance rather than lost work.
Budget for change. Treat AI tooling as a variable cost. Set a monthly ceiling per engineer and review it quarterly, because vendors are adjusting plans every few weeks.
The bottom line
Pro $200 is available again, but it buys about half the API-equivalent usage it did. In exchange, OpenAI promises no 5-hour cap, falling model prices and some extras yet to be announced. For most small teams, the sensible response is to measure real usage per seat, compare it with API pricing, and keep a second tool ready. The past month showed that access to OpenAI's top tier depends on how much compute it has spare.