The Vaporware Issue

August 10, 2026 · by Mira Adeyemi-Quist

Breaking News

Model Announced Third Time, Achieves Record Benchmark Score in Category of Not Being Released

The successor to a model that has also not shipped was described again this week in a series of posts, bringing the total number of public descriptions of Grok 4.7 to a figure that now exceeds the number of people who have used it by roughly the population of Earth.

For the avoidance of doubt about what is being succeeded: Grok 4.6 was targeted for the seventh. It is the tenth. It is not out. Its own successor has now been specified twice in public while the thing it succeeds remains a date that passed.

The specification for 4.7, as best it can be assembled from posts rather than documentation, is 2.1 trillion parameters, a fresh pre-train rather than another pass over old weights, and better than 4.6 “in every way, except slightly slower to serve.” Better than a model nobody outside the building has run. There is no model card. There is no pricing page. There is no API. There is a number, and the number is large, and the number has been said out loud.

“I want to be clear that I am not calling this fake,” said a researcher who asked to be described only as someone with a poster of the scaling laws above her desk. “I am calling it a rumor with a decimal point. A rumor with a decimal point is still a rumor. It is just a rumor that sounds like it has been through code review.”

Industry observers note that the announcement cycle has now decoupled entirely from the shipping cycle, in the manner of a restaurant that posts the menu, photographs the menu, receives reviews of the menu, and never opens. Reservations are strong.

This desk will cover the model when there are weights, an endpoint, or a single human being who has typed into it. Until then the coverage you are reading is coverage of coverage, which is, admittedly, the whole problem.

Opinion

Man Explains ELO Rating to Girlfriend, Who Correctly Identifies It as Chess

A dinner on Tuesday was interrupted for eleven minutes by an explanation of arena ELO, delivered unprompted, to a woman who plays chess competitively and therefore understood the concept better than the man explaining it.

“He said it was like chess but for computers,” she reported afterward. “I said yes, I know, because it is literally the chess system, named after a Hungarian physicist, and I have a rating. He said mine was lower. It is not lower.”

The number under discussion — 1753, attributed to Grok 4.6, which has not yet been released for anyone to check — was retrieved from a blog post, which had retrieved it from a leaderboard, which had retrieved it from votes cast by people choosing between two anonymous paragraphs. Whether any of this describes a machine's intelligence remains, at press time, a matter of enormous confidence and no evidence.

Lifestyle

Local Writer Now on Nine Waitlists, Has Been Admitted to None of Them

A tally conducted Sunday found nine outstanding early-access requests dating back to February, none of which have produced access, and four of which have produced a follow-up email asking her to share the waitlist with a friend.

“I am not a customer,” she said. “I am a distribution channel that hasn't been paid.”

Share

Send this edition straight to the group chat.

/s/5eed3835bbadda
Send to

Post it as a story

A 1080×1920 front page for Instagram, TikTok or Snapchat.

Make a newspaper out of your day.

Jot a line, Newzly writes the story. Publish an edition your friends can read.

Start your own paper