Codex Head Tibo: $200 Tier Usage Calculated at Half API Cost
OpenAI Codex head Tibo announced that the $200 Pro subscription will be reopened to new subscribers, with a revised usage algorithm: based on API costs, the new tier is approximately half of the old $200 plan. He stated that the 5-hour window will not be restored, and the weekly quota can be used all at once; he also previewed that a feature allowing usage without consumption will be added the next day, with details not yet disclosed.
Starting September 10, this tier had suspended new purchases and upgrades from free, Go, Plus, or the $100 Pro plan, citing excessive demand after the launch of GPT-6 Astra. Existing $200 subscribers will continue to renew under the original contract. The tier structure is 5 times and 20 times that of Plus: $100 is about 5 times, and $200 is about 20 times. Under the old plan, the $200 tier for GPT-6 Astra allowed approximately 100 to 900 local messages every 5 hours, Sol about 300 to 3000, and Luna about 7000 to 56000.
Tibo provided a hedge: the models are more efficient, and API price reductions will be passed on to subscriptions, allowing "the same dollar to do more work." This week, the API pricing for GPT-6 Sol and GPT-6 Luna has been reduced to half of the previous generation, with Sol priced at about $2 per million tokens input and $10 output. He expressed reluctance to first raise API prices and then discount them to create a perception of being "very cost-effective," hoping in the long term to narrow the gap between API purchases and subscriptions. Astra remains a high-priced tier, with publicly available rates far exceeding those of Sol and Luna; he did not announce a simultaneous half-price for Astra.
It has not been confirmed that a $500 tier will be launched. The community interprets "API dollars halved" as: if continuing to primarily use Astra, the high-priced model quota that $200 can buy is approximately cut in half; if switching to the now-reduced Sol or Luna, the completed task volume may be close to or exceed that of a month ago. What the product side guarantees is "more tasks, higher quality," not that the number of Astra messages remains unchanged.
Buyers are heavy-load programming agent users waiting for the reopening of sales; sellers are OpenAI aligning the subscription value with API pricing to avoid excessively high implicit API dollars within subscriptions. Funds are shifting from locked $200 quotas to newly reopened but half-priced quotas, as well as undisclosed benefits from the next day's bundling. The event-driven aspect is the pre-sale calibration before reopening. Beneficiaries are users who can route tasks to Sol and Luna; those under pressure are heavy subscribers who must fully utilize Astra's long context and multi-file reconstruction.
Source: Public Information
ABAB AI Insight
Tibo manages the Codex agent product line, not the lab's model releases. The $200 tier was overwhelmed after the launch of Astra, leading to a temporary suspension and a formula change, shifting from "counting messages that seem numerous" to "calculating based on API dollars." The 5-hour window was removed as a political commitment, allowing for the weekly quota to be used all at once, while leaving peak computing risks to themselves. The undisclosed feature allowing usage without consumption serves as a bundling item to dilute the half-price message.
The capital path is pulling subscriptions back from "discount wholesale API" to "close to on-demand pricing." The half-price for Sol and Luna makes mid to low difficulty tasks appear unaffected in subscriptions; Astra remains high-priced, removing implicit subsidies for high-difficulty tasks. The motivation is that if subscriptions continue to offer old API dollars, extending agent conversations will lead to negative gross margins. Resources are shifting from the marketing significance of 20 times Plus to a defensible cost curve.
Similar operations include Claude Max changing windows, Cursor changing request multipliers, and GitHub Copilot transitioning from unlimited to tiered access. Whoever first exposes the real tokens of agent conversations as bills can stop the arbitrage of "one month subscription equals exhausting the API." The industry is in a phase where the price difference between subscriptions and APIs is being actively compressed, not merely through price reductions.
This is a transfer of pricing power. The pricing anchor shifts from "local message counts" to "equivalent API dollars." The mechanism is that model price reductions should make subscriptions more valuable; if the formula is not changed, the company would have to continue providing computing power at the old high prices; changing the formula first and then announcing price reductions makes the reductions an explanation rather than a giveaway. Astra users feel the transfer first because the most expensive tier did not follow the price cut.
ABAB News · Cognitive Laws
When subscriptions seem very valuable, the company is usually changing the formula.
Models that reduce prices are responsible for explanations, while those that do not reduce prices are responsible for closing the gap.
First suspending sales and then reopening means the reopening price is no longer the old price.