The post
On Thursday 3 September 2026, @OpenAI posted three sentences and a video:
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.
Read at 16:06 UTC the following day, that post had 88,570,155 views, 279,037 likes, 28,488 reposts, 21,344 quotes, 6,860 replies and 86,704 bookmarks. It is the largest single launch post in our directory, which currently holds 1981 launches measured on X.
Every figure here is a reading taken at a moment, and this one is still moving fast: the same post was at 37.9M views at 05:32 that morning and 67.9M at 11:51, so it added 20 million in the four hours before we wrote this down. Treat the ratios as the durable part and the totals as a snapshot.
The model itself went out as a limited preview for trusted partners, with the wider rollout to Plus, Pro, Business and Enterprise, the API and AWS set for 5 September at $10 per million input tokens and $50 per million output. On the day of the post, almost nobody reading it could open the thing.
What 88.6M looks like against everything else
The median launch we have measured did 324,948 views. Astra did 273 times that. It accounts for roughly eleven percent of all the views across every launch in the directory combined, which is one post carrying a ninth of a year of AI launches.
The obvious explanation is that OpenAI is enormous, and it only half works. OpenAI has 5.2 million followers, so the post reached seventeen times its own audience. That is a good multiple, and two of the four launches sitting beneath it did better:
| Launch | Views | Followers | Multiple |
|---|---|---|---|
| OpenAI, GPT-6 Astra | 88,570,155 | 5,218,360 | 17.0x |
| xAI | 48,159,946 | 214,306 | 224.7x |
| SpaceX | 44,277,509 | 41,900,168 | 1.1x |
| XBOX | 28,277,114 | 264,630 | 106.9x |
| Anthropic | 19,843,567 | 1,767,429 | 11.2x |
xAI reached 225 times its following. SpaceX, with 41.9 million followers, barely cleared its own audience. Size of list and size of launch are loosely related at best, which is the whole argument for treating distribution as something you build rather than something you already have.
Four rates, and the one that went the other way
Absolute numbers flatter the biggest account in the category. Rates per ten thousand views are the fairer comparison, and they say something more interesting.
| Signal | Astra | Directory median |
|---|---|---|
| Likes per 10K views | 31.5 | 24.4 |
| Reposts per 10K views | 3.2 | 2.3 |
| Quotes per reply | 3.1 | 0.4 |
| Bookmarks per 10K views | 9.8 | 11.5 |
Three of the four sit above the median and one sits below it. The three above are all public signals: liking, reposting and quoting are things you do where other people can see you do them. The one below is the private one.
Bookmarking is the closest thing X has to intent. Saving a post has no social return, so people do it when they expect to come back to something.
At 9.8 per ten thousand views against a median of 11.5, the biggest launch we have ever measured converted attention to intent about fifteen percent worse than an average one.
That number fell as the post grew, from 10.7 six hours earlier, so the marginal viewer arriving late was worth less than the one who arrived early.
Three quotes for every reply
The quote ratio is the number to sit with. The median launch gets 0.4 quotes per reply, so most launches are conversations: people answer the company.
Astra got 3.1. Most of that audience was talking about OpenAI to its own followers rather than to OpenAI.
That is what a launch looks like when it stops being a product announcement and becomes a topic. It is the most efficient distribution mechanism on the platform, because every quote is a new post with its own reach, and it is the mechanism that produced most of those 88.6 million views.
It also explains the bookmark rate. A quote-driven launch spreads through commentary, and commentary reaches people who are interested in the industry rather than people who are shopping for a tool.
Reach scaled 273 times. Reach and demand are different things, and only one of them scaled.
The part that was avoidable
Astra was announced two days before most people could use it. That is a defensible choice for a frontier lab managing a staged rollout, and it has a measurable cost: the single largest audience ever assembled for an AI launch arrived at a page with nothing to do on it.
A launch post carries attention for something like forty-eight hours. Spending that window on a product in limited preview means the people who cared most had to remember to come back, and remembering is where funnels die. The bookmark rate is what that looks like in the data.
If you are shipping to a waitlist rather than to production, the announcement is worth holding until the thing opens. The attention does not wait for you.
What happened to everyone else
Thirteen other companies launched on X on 3 September, to a median of 159,156 views. Thirteen launched the day before, to a median of 312,971. Same number of companies, roughly half the reach each.
Figure AI cleared 1.5 million and Hook cleared 1.1 million. Below those two, the day thinned out quickly.
One caution before you lean on that. Both cohorts were read at the same moment, so the 3 September posts are a day younger and still earlier in their climb. Astra itself added twenty million views in four hours, which is the scale of movement still available to a young post.
The launch directory carries the current numbers, and the honest version of this comparison is that launching alongside Astra cost something material rather than that it cost exactly half.
The direction matches what the platform does. Attention on X is winner-take-most within a topic on any given day.
When the biggest account in your category takes 88 million views on a Thursday, the recommendation system has been handed an enormous amount of evidence about what people in your category want to see, and it is not your seed round.
What to do with a day like that
Move. A launch date is a variable, and moving it costs a week at most. Halving your reach costs the launch.
Frontier labs telegraph their releases through developer events, conference keynotes and staged preview access, usually with more than a week of warning, so the calendar is readable if someone is reading it.
Or ride it. The alternative to competing with a topic is joining it. Quotes are where the reach is on a day like this, and a genuinely useful take on what a computer-use model changes for your category will travel further attached to Astra than it would on its own.
This works when you have something specific to say and fails badly when you are visibly attaching yourself to the news.
Do not split the difference. Posting your own announcement into the middle of someone else's launch day is the one option with no upside. You spend the post, you spend the amplification, and the numbers come back looking like a Tuesday in August.
The wider point holds beyond one Thursday. Reach at this scale is available to a company with 214,000 followers and out of reach for one with 41.9 million, depending entirely on whether anyone chose to carry the launch. That choice is the thing worth engineering.
Questions we get asked
How many views did the GPT-6 Astra launch post get?
88,570,155 views when we read it at 16:06 UTC on 4 September 2026, alongside 279,037 likes, 28,488 reposts, 21,344 quotes, 6,860 replies and 86,704 bookmarks. It is the largest launch post in our directory. The post was still climbing quickly when we read it, adding twenty million views in the four hours beforehand, so the launch page carries the current reading.
Was the GPT-6 Astra launch successful?
On reach, comfortably. It did 273 times the median launch we have measured and roughly eleven percent of all views across the entire directory.
On intent it came in under par: 9.8 bookmarks per ten thousand views against a median of 11.5. The audience was the largest we have recorded and it saved the post at about fifteen percent below the normal rate.
Should I launch on the same day as OpenAI?
No. The thirteen companies that launched alongside Astra on 3 September had a median of 159,156 views, against 312,971 for the thirteen that launched the day before.
Move your date or attach yourself to the topic with something worth reading. Posting your own announcement into the middle of it is the one approach that reliably wastes the launch.
Why did such a big launch have a below-average save rate?
Two reasons that reinforce each other. It spread through quotes rather than replies, at 3.1 quotes per reply against a median of 0.4, so most of the audience arrived through industry commentary rather than through interest in buying a tool. And the model was in limited preview on the day, with the wider rollout two days later, so the people who did want it had nothing to open.