OpenAI has put GPT-6 Astra into its paid ChatGPT tiers, and the most precise thing published about the model is how little of it a customer gets. Astra draws on the same subscription quotas as before, and the standard version consumes them roughly twice as fast: in each five-hour window a subscriber gets about half the messages GPT-5.6 Sol allowed, a ratio that holds on every plan. Free and Go subscribers get no access to either frontier model at all.
GPT-6 Astra Pro is limited to Pro, Business and Enterprise subscriptions, where its use counts against the allowance the plan already had:
Pro at $200 — 200 messages a week for GPT-6 Pro.
Pro at $100 — 50 messages a week.
Business Premium — 50 messages a week.
Business Standard — 15 messages a month.
For the standard model the figures are given as ranges. Plus users can expect between 5 and 45 messages with Astra, against 10 to 100 with Sol. On the Pro 20x plan it is 100 to 900 with Astra against 200 to 2,000 with Sol. OpenAI says the number a customer actually sees depends on the model selected, the length of the context, the reasoning mode, tool use and caching.
Those ranges deserve a second look. A quota quoted as "5 to 45" is not a quota; it is a ninefold spread that resolves only after the work is done. Nobody can plan a working day around a number that swings by a factor of nine depending on how long their prompts run and whether the cache happens to hit. The same is true at the top: 100 to 900 on Pro 20x is the same spread with more zeroes. What OpenAI has published is not a limit but a distribution, and the customer carries the variance.
The Business Standard line is the one that gives the structure away. Fifteen messages a month is fewer than one per working day. That is not a seat for doing work with a frontier model; it is an allowance for finding out what the frontier model is like, priced so that a team which decides it needs Astra has to move up a tier to get it. The tier table is the product here as much as the model is.
The halving is the other signal. Cutting the message budget in half across every plan, uniformly, is a statement about what Astra costs to serve rather than about what any particular customer is doing. OpenAI has effectively told its paying users that its newest frontier model is about twice as expensive per interaction as the one they were using last week, and has passed that straight through as scarcity instead of price.
Notably absent from all of this is any measure of what Astra does better. The only quantity attached to the difference between OpenAI's new frontier model and its previous one is a ratio of message limits — half as many, everywhere. A launch that describes a model entirely through its rationing is a launch making an argument about supply, not about capability.
The frontier has become a metered utility, sold in messages rather than tokens, rationed tier by tier, and withheld entirely from the two cheapest plans. For the subscriber who pays for it, the upgrade that arrived this week is a smaller number.