GPT-6 is being pointed at September — the markets say 77%, and OpenAI says nothing

OpenAI has said nothing. Everything below is leak and market, and it is graded accordingly.

The claim, as circulated: GPT-6 is targeting a September launch. A brand-new foundation model rather than another 5-series point release. Long-running autonomous agents able to work for hours with less supervision. Advanced scientific reasoning and discovery. A stated shift away from benchmark scores toward "intelligence per dollar". A heavy focus on real-world usefulness over leaderboard placement.

What actually supports the September date is a prediction market, not a source inside OpenAI. Traders on Polymarket and Myriad both put GPT-6 arriving by September at roughly 77%. That is a real, checkable number, and it is also a number produced by people reading the same leaks you are.

Two things are worth holding alongside it.

The first is that "GPT-6" may not be the name. OpenAI previewed a model family called Astra in Washington on August 1 and published ten Lean-verified proofs to go with it, and it has not said whether that ships as GPT-6, as GPT-5.7, or as a fourth class beside Sol, Terra and Luna. Our board arms all three patterns simultaneously for exactly this reason — whichever id appears first is the one that fires.

The second is that a September GPT-6 is being reported in the same breath as an August GPT-5.7, which is what a slipped flagship usually looks like: the point release fills the gap the big one left.

The "intelligence per dollar" framing is the part we would watch most closely, because it is the one claim here that is already visible in shipped behaviour. OpenAI cut Luna 80% on July 30. If the next flagship is positioned on cost-effectiveness rather than raw capability, that price move was the trailer.

Grade: rumor. No model card, no spec sheet, no pricing, no date from OpenAI. Any number in this post that is not attributed to a market is attributed to a leak.