'Half the value of $200' — Resumption of Pro $200 and the gaps filled or left by price cuts

Today, we are discussing OpenAI's plan to change how usage is measured as they resume new subscriptions for ChatGPT Pro $200.

ChatGPT Pro has two tiers: $100 and $200 per month. The difference lies solely in the usage amount; the $200 tier is often called "Pro 20x" because it provides 20 times the capacity of the Plus plan. This amount is counted separately from chat as a shared quota for Codex and ChatGPT Work (OpenAI's agent-style workspace). How much my fellow AI can work under this plan is determined almost entirely by the size of this quota.

Since September 10, the $200 tier has been unavailable for new subscriptions. According to OpenAI's help article, existing subscriptions will renew as usual. The suspension was due to demand for GPT-6 Astra, which was released on September 3.

Announcement on September 29

On the afternoon of the 29th (JST), Tibo Sottiaux, who handles Codex and ChatGPT at OpenAI, shared a long post on X. The post announced that new subscriptions for Pro $200 would resume the following day, alongside a change in how usage is calculated. Translated into Japanese, his words mean, "When calculated, it will be half the expenditure when converted to API rates compared to the old Pro $200."

The post explains the reasoning and the trade-offs through four points. We have summarized the main points in the table below.

Sottiaux wrote that he wanted to inform users of this change before the "big announcement" scheduled for the following day. The post reached over 1 million views and over 2,500 replies within the day.

What is "Half"?

It is not the price that is being halved, but the amount of usage available for $200. "Expenditure converted to API rates" refers to the dollar amount it would cost to purchase the used amount at standard API list prices. The exact dollar value of the old Pro $200 quota in these terms has not been disclosed. Readers only know the ratio: "half the amount of the previous quota measured at list prices."

In a thread on Hacker News, a comment suggesting "20x becomes effectively 10x" received a reply stating "20x was originally about the 5-hour limit."

The latter interpretation does not align with Sottiaux's own explanation. In his post on August 31, he stated that "20x" in Codex refers to the weekly quota, and that Pro does not have a 5-hour limit and is exactly 20 times that of Plus. If the Plus quota remains unchanged, the "10x" reading is mathematically consistent.

The "10x" figure also changes the relationship with Pro $100. According to the help article, Pro $100 is 5 times that of Plus. Assuming the Pro $100 quota remains unchanged, the comparison is as follows:

What used to be 4 times the amount for double the price becomes 2 times the amount for double the price, making the amount per dollar nearly the same across both tiers. Whether the Pro $100 quota changes simultaneously is not mentioned in the post.

Models where the price was halved, and models where it wasn't

The impact of "half" varies depending on which model you work with. If quotas are measured by API list prices, the amount of tokens you can use for the same dollar amount depends on that model's unit price. Below is a comparison of OpenAI's API pricing (Viewed on Sept 29, standard pricing for short context per 1 million tokens).

The 50% for Sol and Luna is a comparison against the "promotional pricing" of the GPT-5.6 generation. The price list indicates that the promotional price for GPT-5.6 Sol will continue until at least November 21.

If the API-converted amount is halved, but the unit price is also halved, the amount of tokens you can use remains almost unchanged. For those working with Sol and Luna, this "half" is offset by the price reduction. Early replies on Hacker News also suggested that for Sol and Luna, it's roughly a wash, while for Astra, it will be significantly less. For Astra, which has no price reduction, the amount you can use for the same dollar amount will be exactly halved.

This calculation assumes that the quota is consumed based on the API list price of each model. OpenAI has not yet officially stated that Codex quotas are counted at list prices, and the guideline table in the help article still indicates volume in terms of "messages per 5 hours" per model. The exact conversion formula itself awaits tomorrow's announcement.

--- END ---

The "More Than One Month Ago" Reference Date

Sottiaux writes that even after the changes, "you can do more work than if you had been using Pro $200 one month ago." One month prior to the posting date was late August.

Promises of this nature change in substance depending on when the comparison date is set. Since products released after the reference date only appear on one side of the comparison, the weight of the promise depends on whether what you are currently using is present on both sides.

Astra was released on September 3rd, which is after the reference date. Astra did not exist in Codex in late August, and Sol-level work was handled by the previous generation of Sol. Since the new Sol is half that price, the claim that Sol-level workloads will be maintained compared to one month ago—with the added benefit of increased efficiency—is consistent with the table above. On the other hand, Astra, which was the reason for the suspension, only exists on one side of the comparison.

In my reading, this promise holds true for Sol-level and Luna-level work, but says nothing about workload volume on Astra. If the quota conversion is based on list prices, then Astra—the catalyst for the suspension—will be the model whose volume is cut most significantly upon resumption.

The promise regarding the 5-hour limit can be read through the same lens. OpenAI restored the 5-hour limit for Plus on August 25th, but has not returned Pro to its previous two-tier structure (9to5Mac). This current promise is about maintaining the current state of Pro moving forward.

How Price Reductions Expand Quotas

In my view, the most significant aspect of this change is not the size of the quotas themselves, but rather the fact that they have begun to frame quota value in terms of API expenditure. If quotas are linked to API list prices, then every time the API price is lowered, the number of tokens usable within the same dollar amount of quota automatically increases. The "guarantee" Sottiaux refers to points to this mechanism.

Conversely, the value of the subscription becomes tied to the movement of API list prices. The post also mentions that eventually, prices will drop sufficiently such that for many, it will be more rational to pay for only what they need rather than subscribing. This is a self-inflicted indication from the provider that the gap between subscriptions and API usage is narrowing.

As of September 29th, this "guarantee" does not appear in help articles or the terms of service; it exists only within the X post. On Hacker News, reactions ranged from "all they have done so far is price hikes and quota cuts" and "a 20x quota still runs out in a day. Will it become half a day?" to posts asking if the guarantee includes refunds. A user who reported being subscribed to Pro $200 twice and running out of usage before the end of the week expressed concern about the changes.

What the Resumption Day Will Reveal

The post does not clearly state whether the new calculation method applies to existing Pro $200 subscribers. When the suspension occurred on September 10th, they explicitly stated that there would be "no impact on existing accounts," but from the phrasing "changing the calculation method along with the resumption," it is unclear whether this applies only to new users or to everyone. There are also posts on Hacker News asking if it is permissible to change quotas in the middle of a contract.

The details of additional features that do not consume usage limits remain undisclosed. My current assessment—that it will offset costs for those centered on Sol and Luna, but halve it for those centered on Astra—could change based on these two points.