The forecast before the forecast
The Kyle Reese Time Machine Appreciation Society
A busy August gave way to an even busier September—and a new AI pastime: ranking models before their makers have announced them. Arena offers real clues, but it does not offer a calendar from the future.

August 2026 was crowded with AI releases. September has found a way to be crowded with releases that haven’t happened. If you’re following the industry closely enough, apparently you can now be late to a conversation about next month’s models.
Welcome to the Kyle Reese Time Machine Appreciation Society. Reese is the soldier who travels back in time in The Terminator and becomes John Connor’s father. Our version of the club has a less urgent mission: returning from the future to explain why a model that has not been announced is supposedly slipping down a leaderboard.
There is a real mechanism beneath the joke. Arena says providers can let its community test unreleased models with their names hidden. In its blind, head-to-head battles, people choose the better of two answers; those preferences help produce public rankings. An unfamiliar label can therefore be worth watching. For a look at how quickly the visible rankings can change, see what happened between August and September.
But the leap from an unfamiliar label to a named product with a score, a launch date and a place in a future lineup is doing almost all the work. Names such as Opus 5.5, Grok 4.7 and Gemini 4 may fit beautifully into a speculative chart; that does not make the chart a company roadmap. The claimed post and screenshot behind one such lineup were not archived in the material available to us, so even that particular list’s provenance remains unverified.
An anonymous model in a blind test is a clue, not a release date.
This is what makes the new game so absorbing. A genuine anonymous preview creates a small pocket of uncertainty. Fans fill it with guesses about identities, capabilities and what might ship next. Before long, a forecast can acquire the visual authority of a benchmark, complete with rows and positions, while still being a forecast.
First we measured the models that launched. Then we watched the leaderboard move. Now we’re ranking the models we think might be waiting just offstage. At this rate, the society’s next dispatch will be a benchmark for a model whose research team hasn’t thought of it yet.
Imagerly.Ai · No. 01 · September 2026