Updated August 26, 2026 · 13 min read · by Jake Hari. Panel verdicts were generated August 26, price-blind, from Kalshi's binding contract terms and a fetched record of what OpenAI and its CEO have actually said. Kalshi prices fetched August 26, 2026, 19:54 UTC.
On Sunday, Sam Altman sat down with David Senra on the Founders podcast and spent part of it explaining what he had gotten wrong. He had been "too ambitious on timelines." The disruption he expected after GPT-4 had not arrived, because "the economy just has so much inertia." He moved his own forecast for superintelligence out to the end of 2028.
By Tuesday night, traders on Kalshi were buying the opposite.
The exchange lists three contracts on whether OpenAI announces it has achieved AGI, one for each of three deadlines. On Monday the nearest of them, the one covering the rest of this calendar year, closed at six cents. By Wednesday afternoon it had traded as high as thirteen and was bid at fifteen. The 2027 contract went from eighteen cents to a twenty-seven-cent print. The long one, running to the end of 2029, touched forty-seven.
Nothing OpenAI published caused that. What happened in between was a rumor, and the rumor is not really the interesting part. The interesting part is that almost everyone trading this market appears to be answering a different question than the one Kalshi wrote down.
Free: The Weekly PM Market Brief — the 8-model panel's graded record, the week's biggest market-vs-model gaps, and what's spiking next. One email, Sundays.
New to event markets? Our plain explanation of how prediction markets work covers the mechanics in about five minutes.
What This Contract Actually Pays On
Read the settlement language and the whole board changes shape.
"If OpenAI announces that they have attained AGI by Dec 31, 2026, then the market resolves to Yes." That is the entire rule for the near contract, verbatim. The other two are identical apart from the year.
There is no benchmark in that sentence. No capability threshold, no scoreboard, no panel of outside scientists. If OpenAI runs some internal review before making a claim like that, the market's settlement language does not know about it and does not wait for it. The contract does not ask whether artificial general intelligence exists, or whether anyone agrees it exists. It asks whether one company publishes a particular claim about itself before a particular date.
Kalshi's binding contract terms for the series make that even sharper. The Underlying, in the exchange's language, is "releases from OpenAI." The Source Agency, the authority whose word settles the market, is listed as: OpenAI.
The company the contract is about is the source whose own releases decide whether it pays.
That is not a flaw Kalshi overlooked. It is the only workable design, because AGI has no agreed measurement and someone has to be the referee. But it means the thing being traded here is a corporate communications decision, and it should be priced like one. Every trader buying this contract because Astra solved ten open math problems, which is the thing OpenAI's unreleased next model did on August 1, is buying a claim about a press release using evidence about a research result.
And it puts unusual weight on a definition. OpenAI's own charter defines AGI as "highly autonomous systems that outperform humans at most economically valuable work." That bar is economic, not cognitive. It is not about whether a model can reason, write proofs or pass an exam. It is about whether the work of the economy has moved.
Which is exactly the bar Altman spent Sunday saying has not been cleared.
The Clause That Used To Make Silence Free
For years there was a hard, boring, financial reason to expect OpenAI never to say those words.
The original 2019 arrangement with Microsoft made Microsoft's commercial rights to OpenAI's technology contingent on an AGI declaration by OpenAI's board. Declaring AGI would move IP and revenue rights. It was, as Simon Willison documented in tracing the clause's history, an undefined threshold one party could call unilaterally, and Microsoft was reported to be worried OpenAI might call it early off a strong model.
That reason is gone.
In October 2025 the "next chapter" amendment routed any declaration through an independent expert panel, so OpenAI could no longer call it alone. Then on April 27, 2026, the amended deal removed the AGI trigger entirely. Microsoft's license became non-exclusive through 2032. OpenAI's revenue share to Microsoft now runs to 2030 independent of OpenAI's technology progress, which is to say regardless of whether anyone declares anything.
Until the end of April, declaring AGI cost OpenAI something specific and expensive under its Microsoft agreement. That particular cost is now gone. If you were relying on the old clause as the reason this market stays near zero, your reason expired in April and you may not have noticed.
That change is real and it points up. It is also, as far as our panel could tell, roughly the only thing on this board that does.
The Board
Here is where the market sits and where six AI models landed after pricing the same three contracts without being shown a single price.
Each model received the binding contract terms, the charter definition, the full documented history of the Microsoft clause, Altman's statements from Sunday, OpenAI's August research announcements, and a coverage note spelling out what it was and was not being shown. None saw a price. Then each read the other five arguments anonymously and got one chance to revise.
| Deadline | Market (bid / ask) | Panel | Gap |
|---|---|---|---|
| Before 2027 (By Dec 31, 2026) | 15¢ / 22¢ | 0.75% | market is ~20x the panel |
| Before 2028 (By Dec 31, 2027) | 23¢ / 30¢ | 8.0% | market is ~3x the panel |
| Before 2030 (By Dec 31, 2029) | 37¢ / 44¢ | 31.5% | market is ~1.2x the panel |
Prices are the live order book as of August 26, 2026, 19:54 UTC. The near contract is the thin one: its spread runs seven cents wide and only 538 contracts changed hands in the previous 24 hours, against 6,673 on the 2027 rung. Read its price as a range, not a number. This board is repricing through a live news wave: the quotes on this page are stamped, and the page is refreshed as the story moves. The panel column is the blended read: the median of the six seats after the revision round, which is what keeps one outlier from dragging the number around.
Every seat's number, before and after the revision round:
| Seat | Before 2027 | Before 2028 | Before 2030 |
|---|---|---|---|
| Claude Fable 5 | 1.2 → 1.2 | 7.0 → 8.0 | 30.0 → 35.0 |
| Claude Opus 5 | 0.8 → 0.8 | 5.5 → 6.0 | 22.0 → 25.0 |
| Claude Sonnet 5 | 0.6 → 0.6 | 5.0 → 5.0 | 28.0 → 28.0 |
| ChatGPT | 0.7 → 0.7 | 18.0 → 12.0 | 54.0 → 46.0 |
| Gemini | 0.4 → 0.5 | 12.2 → 8.0 | 58.5 → 32.0 |
| Kimi | 2.0 → 2.0 | 9.0 → 9.0 | 28.0 → 31.0 |
| Panel Middle | 0.75 → 0.75 | 8.0 → 8.0 | 29.0 → 31.5 |
Six of the panel's eight seats ran this board, and the middle is taken across the six that answered.
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 84% on 761 graded calls · Claude Opus 84% on 831 graded calls · Claude Sonnet 82% on 808 graded calls · GPT 84% on 6,249 graded calls · Gemini 86% on 4,952 graded calls · Kimi 83% on 2,534 graded calls. Recomputed daily; the full scoreboard is public.
These are model estimates, not predictions of fact and not financial advice. Kalshi is a CFTC-regulated exchange for event contracts, 18+ only, and availability varies. Every number in this piece gets graded in public once the market settles: the running record lives on the full graded scoreboard.
The near rung is the one worth staring at. Every seat put it at or below 2%, and four of the six put it under 1%. The market is bid at 15¢. The seats all got there by the same route: a company is not going to announce it has met its own economic definition of AGI in the same calendar year its chief executive publicly explains that the economic effects have not shown up.
There is a second check available, and it does not flatter the near contract either. Kalshi runs a separate ladder asking whether any company announces AGI, with stricter wording that explicitly rules out "AGI-like," "AGI-level" and "near-AGI" claims. On that board, any company at all doing it before January 1, 2027 is bid at eleven cents, fetched in the same pass as the prices above. OpenAI is one company inside that field, and its own contract is bid four cents higher than the whole field.
There are only two ways that can make sense. Either one of the two boards is mispriced, or traders are pricing the wording gap: OpenAI's contract says only that it must announce it has "attained AGI," while the any-company contract spells out that "AGI-like," "AGI-level" and "near-AGI" do not count. A hedged claim in an OpenAI launch post is a live argument for settling the first board and a dead one for settling the second. If that is what the four cents is, it is a bet on how loosely one sentence gets read, not on whether AGI arrives.
More live boards from the same panel: OpenAI's IPO board, where a June 2027 announcement is bid at 71¢ and CFO Sarah Friar told staff on August 19 that OpenAI "will be a public company in 2027" · the top-ranked AI model board, where OpenAI leads the field at 25¢ bid · and GPT-6 release date odds, where before-2027 is bid at 67¢. Prices fetched August 26, 2026.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $194M traded
- 2027 Pro Football Champion · $72M traded
- 2028 U.S. Presidential Election winner? · $62M traded
- Pro Baseball Champion · $61M traded
- 2028 Republican presidential nominee · $60M traded
- Bitcoin price at the end of 2026 · $33M traded
Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.
Where The Panel Changed Its Mind
The revision round barely touched the two near rungs and pulled the far one from a 36-point spread down to a 21-point one. Two seats came a long way to get there.
The largest single move on the board belonged to the seat that started highest, and it moved because it changed its mind about a piece of corporate plumbing.
"I am revising my 2030 rung down sharply, persuaded by Forecaster C's assumption that the expert panel likely died alongside the Microsoft trigger it was meant to verify. Without that institutional forcing function, Forecaster D's argument about Altman's 'smooth curve' framing becomes the dominant barrier to a discrete announcement." — Gemini, second round
Its reasoning is worth following, because it inverts the obvious read. In round one it had assumed the October 2025 expert panel survived, and treated that as a path: a formal verification process gives a company a structured, defensible way to make a declaration. Once it priced the assumption that the panel died with the clause it existed to police, the declaration stopped being a process anyone would walk through and became "an entirely unforced corporate communications choice." Unforced choices with no upside do not get made. Its 2030 number fell from 58.5% to 32%.
[Editor's note: the public record establishes that the April 2026 amendment removed the AGI trigger. It does not say what became of the expert panel that the October 2025 amendment created to verify it. Four of the six seats priced the panel as having lapsed with the clause it existed to police. That is their reading, not a documented fact, and it is the single assumption doing the most work on the long rung.]
The move in the other direction came from the seat that started lowest.
"I nudged the long rung 22→25 for one channel I underweighted: a rival declaration as a forcing function. If any company says the words first, OpenAI — which coined the charter definition and is the frontier lab — is the most likely fast follower." — Claude Opus 5, second round
One seat refused to move at all, and did so by attacking the arithmetic everyone above it was using. The move it objects to is a simple one: take the odds that some company says the words, then assume OpenAI is the one that says them roughly four times out of five.
"Holding against Forecasters B and C: they split the any-company mass giving OpenAI ~77-83% of it, but the definition trap binds OpenAI more tightly than rivals since OpenAI itself coined the economic bar and its own CEO just admitted it's unmet — that argues for a LOWER OpenAI share of the any-company mass, not a higher one. I stay at 28/5/0.6, an outlier by design, not by oversight." — Claude Sonnet 5, second round
That is the sharpest idea on the page. Every other lab can use the word "AGI" loosely, in a launch video, without contradicting a document it wrote. OpenAI wrote the document. The company with the most credible claim to getting there first is also the one whose own definition makes it hardest to announce.
What Would Change The Panel's Mind
Four triggers, all of them dated or checkable rather than atmospheric.
Astra ships publicly. Kalshi has a separate contract on it, and traders bid a release before October 1 at 75¢, and before November 1 at 86¢. Astra is the model that produced ten open-problem proofs in August while still unreleased. A public launch is the most likely occasion for OpenAI to attach big language to a product this year. Direction: pushes the near rungs up, and it is the only thing on this list that can plausibly move the 2026 contract at all.
Any rival says it first. This channel produced the panel's largest upward revision. The seats' reasoning is that if Google, Anthropic, xAI or Meta declares AGI, OpenAI's incentive flips from "why would we" to "we cannot let that stand." Watch Kalshi's any-company ladder as the tell, not the news. Direction: pushes every rung up, hardest on 2028 and 2030.
Altman's end-of-2028 superintelligence claim comes due. He put it on the record on Sunday. If he stands behind it as the date approaches, declining to use the smaller word "AGI" through all of 2029 becomes difficult to sustain. Two seats named this as the specific reason the 2030 rung is not lower. Direction: pushes the 2030 rung up, and does nothing to the near ones.
OpenAI restates or retires the charter definition. The economic bar is the binding constraint on every near-term rung. If OpenAI publishes a revised definition, or formally moves its public goal to superintelligence and stops treating AGI as a milestone it will ever claim, the near contracts should collapse rather than rise. Direction: down, sharply, on all three.
Settlement timeline
| Date | What happens |
|---|---|
| October 1, 2026 | Kalshi's Astra release contract is bid at 75¢ for a public launch by this date |
| December 31, 2026 | Deadline for the "Before 2027" contract; trading closes 11:59 p.m. ET |
| December 31, 2027 | Deadline for the "Before 2028" contract |
| December 31, 2029 | Deadline for the "Before 2030" contract |
| Any Qualifying Announcement | Each rung whose deadline has not passed settles early, at the next 10:00 a.m. ET. An announcement by Dec 31, 2026 would settle all three Yes |
This page is re-scored when the story moves, and all numbers on it are stamped August 26, 2026.
- OpenAI IPO Odds: The Board Is Trading The Month
- Top-Ranked AI Model 2026: OpenAI Leads The Board
- Best AI Coding Model Odds: One Column Decides It
- Tech Layoffs 2026 Odds: Kalshi's Bet Is Half Won
- How Do Prediction Markets Work? A Plain Explanation
The Bottom Line
The rumor that appears to have moved this market, on Tuesday evening, was that OpenAI has finished pretraining a very large model. Wccftech, which picked it up from an X account, called it "a wild rumor that is currently doing the rounds on social media." OpenAI has not confirmed it. Whether it is true barely matters here, because this contract does not pay on model size. It pays on a sentence.
The gap between the market and the panel is really a gap between two questions. Traders are answering "is OpenAI close to AGI," and on that question a very large unreleased model and ten machine-verified proofs are real evidence pointing one way. The contract asks "will OpenAI say it has AGI," and on that question the evidence runs the other way: the definition it would have to satisfy is one it wrote and its CEO just called unmet, the goalpost has already been relabeled superintelligence, and the framing Altman now uses, a smooth curve with no launch moment, is a framing with no announcement in it anywhere.
One thing did change in OpenAI's favor, and it is the reason the long rung sits near a third rather than near zero. Saying the words used to cost real money. Since April it costs nothing. Nobody has yet produced a reason it earns anything either, and until someone does, a market bidding fifteen cents on it happening in the next four months is pricing a technology story onto a press-release contract.
