The crowd tripled this contract's price in a day, from the 28 cents it traded at through the morning of Thursday, September 10, to 84 by Friday morning, on a rumor and a sentence, and neither one can settle it. On Wednesday night, September 9, this page quoted Kalshi's "another Millennium Prize solution before 2027" contract at 34 cents bid, 39 asked, with one contract behind the bid. By Friday morning, September 11, it was 84 bid, 85 asked, the price in this page's title. As of 2:07 a.m. ET on Saturday, September 19, it is 62 bid, 66 asked, with 4 contracts on the bid and 48 on the offer, after a week in which it printed as low as 32 and as high as 84 without a statement from OpenAI. Nothing that settles it has happened. OpenAI has not announced a second solution. What the week's tape and a Thursday report in The Information did to the price is right below the board.
| Kalshi contract | Midpoint | What it says |
|---|---|---|
| 2027 Sep 9 | 36.5¢ | Another solution before Jan 1, 2027, at 34 bid and 39 ask on Wednesday, September 9. |
| 2027 Sep11 | 84.5¢ | The same contract at 84 bid and 85 ask by Friday morning, September 11. |
| 2027 Sep19 | 64¢ | The same contract at 62 bid and 66 ask at 2:07 a.m. ET on Saturday, September 19. |
| 2028 Sep 9 | 50.5¢ | Another solution before Jan 1, 2028, at 50 bid and 51 ask on September 9. |
| 2028 Sep11 | 91.5¢ | The 2028 contract at 91 bid and 92 ask on September 11. |
| 2028 Sep19 | 80¢ | The 2028 contract at 76 bid and 84 ask on September 19, with no trade printed since September 16. |
Midpoints of the bid and ask quoted in the board below; the chart at the top of this page plots the September 9 to 11 rows, the run-up this page was first written about, and the full order-book detail is in the board.
| Contract | Yes bid / ask, Sept 19 | Resting at the best bid / best ask | Yes bid / ask, Sept 11 | Yes bid / ask, Sept 9 | Contracts traded since open / still open |
|---|---|---|---|---|---|
| Announces Another Solution Before Jan 1, 2027 | 62¢ / 66¢ | 4 / 48 | 84¢ / 85¢ | 34¢ / 39¢ | 32,187 / 13,018 |
| Announces Another Solution Before Jan 1, 2028 | 76¢ / 84¢ | 200 / 100 | 91¢ / 92¢ | 50¢ / 51¢ | 9,371 / 3,233 |
Order book as of September 19, 2026, 2:07 a.m. ET; the September 11 column is this page's prior pull at 6:02 a.m. ET that morning, and the September 9 column its first pull at 9:13 p.m. ET that night. Both contracts opened at 6:00 p.m. ET on September 8. Counts are contracts at one dollar of face value, not dollars of volume; "still open" is open interest. The 2028 contract's last trade was 0.06 of a contract at 82 cents at 12:07 a.m. ET on September 16, so its bid and ask are the only live numbers it has. Re-run every number below at the price you are actually offered.
What Moved Since September 16
This page last changed on September 15, and its prices dated from the September 11 pull, so this section covers the whole week since that Friday morning, not just the last three days. The short version: no statement from OpenAI moved the price, the price moved anyway, and the one thing that resembled news arrived on Thursday after the tape had already done most of its work.
| Day (ET) | Contracts traded | Bought / sold by the aggressor | High / low print | Last print |
|---|---|---|---|---|
| Fri, Sept 11 (from 6:02 a.m.) | 2,316 | 617 / 1,699 | 84 / 70 | 79 |
| Sat, Sept 12 | 7,123 | 1,963 / 5,160 | 84 / 37 | 49 |
| Sun, Sept 13 | 3,990 | 3,525 / 465 | 82 / 48 | 82 |
| Mon, Sept 14 | 1,142 | 33 / 1,109 | 65 / 56 | 56 |
| Tue, Sept 15 | 819 | 185 / 634 | 53 / 48 | 51 |
| Wed, Sept 16 | 401 | 6 / 395 | 51 / 50 | 51 |
| Thu, Sept 17 | 1,473 | 754 / 719 | 76 / 52 | 72 |
| Fri, Sept 18 | 536 | 29 / 507 | 71 / 32 | 65 |
The before-2027 contract's per-trade tape, Kalshi, September 11 6:02 a.m. ET through the September 19 fetch; "bought" and "sold" are the side that crossed the spread. Daily totals sum every fill in the day, rounded to whole contracts.
The give-back started within the hour. On Friday, September 11, 2,316 contracts traded and 1,699 of them were sold by the aggressor, the first fill a 15-contract sale at 84 at 7:05 a.m. Saturday was the week's busiest day, 7,123 contracts, with 5,160 sold against 1,963 bought. At 5:23:30 p.m. one second cleared 1,614 contracts at 70 and 71, and three minutes later, at 5:26:44, a second sweep of 810 contracts found no bid at all between 70 and 45 and ran down to 37, the low until Friday's hole print. Buyers answered within two hours, paying 46 and 47 for 1,000 contracts in a single second at 7:05 p.m., and Saturday's last print was 49. The mid-40s buyers the September 11 edition described, about a thousand contracts waiting for the price to come back, got their price by Saturday dinner.
Sunday reversed it. September 13 ran 3,990 contracts with 3,525 bought, and at 6:03:07 p.m. one second cleared 2,493 contracts across 19 fills from 60 up to 82, a size-weighted 77 cents, with 1,167 contracts at 82 alone. That is the largest single second on this contract's whole tape, bigger than any of the September 10 sweeps described in the next section, and it was the day's last trade. Kalshi's daily candle closed Sunday at 82 with the book at 59 bid and 80 asked, a 21-cent spread at the close, which is a price with nobody standing behind it. Monday sellers proved the point: 1,109 of Monday's 1,142 contracts were sold, and after a 65 high on the morning's few buys the day's last print was 56. Tuesday traded 48 to 53, Wednesday saw four fills between 50 and 51, including 395 contracts sold at 50 at 3:09 p.m. In five days the contract had gone 84, 37, 82, 50, and OpenAI had said nothing.
Thursday, September 17, is the day that needs the most care. At 10:18:09 a.m. a buyer took 330 contracts from 52 to 54. At 12:27:52 p.m. another 241 went off from 59 to 63, and at 1:27 p.m. the week's first 75-cent print landed, 33 contracts. The Information reported that day, under Stephanie Palazzolo's byline and citing a person familiar with the matter, that OpenAI employees expect the Hodge Conjecture to be solved relatively soon, and that the announcement could take longer because the company is working out how to make it without another fight with mathematicians; Gizmodo's write-up puts it this way: "It could take the company longer to announce the solution, though, because it's trying to figure out how to collaborate with the math community to make the announcement without triggering another PR nightmare." The earliest relay of that report we can time is an X post at 11:14 a.m. ET; the article's own publish minute is behind The Information's paywall, and the tape's first buying came an hour before the relay, so we will not say the report caused the 10:18 a.m. fills. Sellers met the 75s: about 615 contracts were sold at 75 and 76 between 1:37 and 3:43 p.m., and Thursday finished with 754 bought against 719 sold, a high print of 76 and a last print of 72. Friday's tape was two sell sweeps and a few small buys. At 1:50 a.m. 237 contracts sold at 70 and 71, and at 3:07:52 p.m. a 271-contract sweep cleared the bids at 68, 67 and 66 and then printed 13 contracts at 32, the same shape as the 1.57-contract 62 in the section below, in reverse: a hole in the book, not a price. Ten contracts bought at 65 at 10:58 p.m. Friday is the last trade on the tape.
The reason an announcement might wait is now on the record, which it was not on September 11. On September 10 OpenAI withdrew as a sponsor of Caltech's Mathathon after an open letter from mathematicians said the event had "advanced a campaign of scientific misinformation about the goals of mathematical research," with OpenAI's Dan Roberts writing, per Gizmodo's September 11 report, "We recognize that the rapid progress of AI in mathematics is disruptive." On September 11, 25 Fields medalists published a declaration titled "A Severe Misalignment of AI in Mathematics" on Terence Tao's blog, which says: "The goals of the AI companies and the goals of the mathematical community are severely misaligned." It names no company. The Information's source describes the delay as a version of that same fight, and "working through how to share these results thoughtfully," the line from OpenAI's September 10 statement quoted below, now has a visible reason to take time. The deadline does not move with it. The contract needs the announcement by 11:59 p.m. ET on December 31, and a company that is being careful about how it announces is a company spending runway.
What did not move: the rulebook, quoted in the next section, is word for word what it was. Open interest nearly doubled, 7,165 to 13,018 contracts, so more money is on both sides of this than a week ago, not less. The inside market went from one cent wide to four, and the 592 contracts that stood behind the 84 bid on September 11 are 4 behind the 62 today. The other venues moved too, and the inversion this page flagged last week has closed; that table is further down. First, how the price got to 84 in the first place, because none of it settles the contract.
Stay ahead of the markets.
Daily insights and expert picks on Kalshi, Polymarket, and what's moving markets.
Free forever. Unsubscribe anytime.
What Moved, And Why It Cannot Settle The Contract
Two things happened between Wednesday night and Friday morning. An X account called leo posted at 4:54 a.m. Thursday that "I am told the Hodge Conjecture is very close to being verified by OpenAI, and that one of OpenAI or Anthropic are also close to solving Birch-Swinnerton-Dyer," adding in the same post, "if true - and it may not be, given the scale of the rumour mill right now." That is the post the day's coverage traced the frenzy to, including a write-up at The News International, which said what this page says: there is no confirmation from OpenAI or Anthropic. An earlier version of the same rumor, posted by Dr Singularity at 2:00 p.m. Wednesday, moved nothing at all.
Then OpenAI gave The New York Times a statement, relayed by OfficeChai from the Times' September 10 article on the Navier-Stokes credit dispute. In full: "In addition, since the completion of Navier-Stokes, we have made substantial progress on another Millennium Prize problem. We are working through how to share these results thoughtfully."
Now the rulebook, which did not change. The contract pays only if OpenAI itself announces a complete solution to one of five named problems: the Riemann Hypothesis, P versus NP, the Hodge Conjecture, Birch and Swinnerton-Dyer, or Yang-Mills. Navier-Stokes is excluded by name, and so is the clause that decides this week:
"Partial results, numerical evidence, or formalization of a previously known solution do NOT count as a complete solution."
"Substantial progress" is, in the contract's own vocabulary, a partial result. It is real evidence about what OpenAI might announce next, and the sections below take it seriously as evidence. It is nothing as a settlement event. Three more doors a casual reader would assume are open are also closed: a proof by an independent ChatGPT user does not count even if OpenAI publicizes it, giving a university team model access does not make OpenAI a collaborator, and "Clay acceptance, peer review, and a prize award are not required," which cuts both ways. The settling event is a press release from one company, naming one problem, saying complete.
Reading the settlement clause before the price is the habit that runs through Kalshi's OpenAI IPO contract, which pays on a regulatory trigger rather than a first trade, and through the AGI board, which pays when OpenAI publishes the words rather than when the thing arrives. Same event, different rulebooks is the standing lesson, and this week it produced a live example.
The Day The Price Tripled
Every time below is Eastern, from Kalshi's per-trade tape, and every "bought" or "sold" is the side that crossed the spread. The tape can count contracts and name the aggressor. It cannot count people or say why anyone traded.
The contract sat in the 20s through the small hours of Thursday, September 10. The real buying started at 8:50 a.m., when 170 contracts went off at 28, followed by 250 in one second at 28 to 34 and 193 more at 35. Then, at 9:48:57 a.m., a fill of 1.57 contracts printed at 62 cents. The September 9 edition of this page described the book above 40 as empty "until 62," and that is exactly the price the next buyer paid once the 39 and 40 offers were gone. The hourly candle Kalshi labels 10:00 a.m. opened at 28 and closed at 62 on 444 contracts, so a chart showing this contract "jumping" 34 cents in an hour is showing a 1.57-contract print into a hollow book. Real money followed at 10:42 and 10:47, 400 contracts bought at 63 to 67.
That first leg ran before the Times statement was visible. The earliest timestamp we hold for the statement circulating is 11:26 a.m. Thursday; we could not pin the minute the Times article itself posted. The second leg lines up cleanly: between 12:19 and 12:34 p.m. about 1,400 contracts were bought from 39 up to 67, including 731 in a single second at 12:34:59, and at 12:59 p.m. the first 75-cent print landed.
Then it stopped being one-directional. At 2:09:58 p.m. a seller swept 629 contracts from 81 down to 53 in one second. Twenty-two minutes later 56 contracts were bought from 87 up to 90, the day's high print and the same kind of thin fill as the morning's 62. Buyers took it back twice, then the evening belonged to sellers: 319 sold at 89, 460 at 86 to 88, 336 at 85 to 86, 288 at 77 to 79, 382 at 73 to 74, and just after midnight 26 contracts sold as low as 60. At 1:30:22 a.m. Friday a buyer swept 608 contracts from 77 to 88 in one second, and nothing traded between 1:46 a.m. and the 6:02 a.m. fetch.
That Thursday's totals: 8,300 contracts across 177 fills, a high of 90 and a low of 22, with 5,071 bought by the aggressor and 3,229 sold. That is buyers over sellers by about three to two, not a day where everyone agreed. A price that round-trips 30 cents in an evening on a two-day-old market is a crowd still arguing, and the 84 was that morning's last word in the argument, not its conclusion. The week that followed, above, said the same thing louder.
The book on September 19 says the same thing in a lower key. Behind the 62 bid, which holds 4 contracts, there are 68 at 61, 400 at 60, 52 at 59 and 200 at 58, then a hole down to 43, and then the crowd that has been waiting all along: 86 at 43, 500 at 39, and about 7,300 contracts resting below 20 cents, including 3,111 at 12. Above the 66 offer sit 200 at 67 and 263 at 68, then a thinning ladder to 268 at 75 and 100 at 80. Two crowds still: one paying the 60s, and a much larger one that will only buy this if it collapses. What a bid-ask spread actually costs is now a four-cent problem at the inside market, and Friday's 32-cent print is what a hole like that costs anyone who has to sell in a hurry; Friday's sweep fell through a 34-cent gap under 66 that the book has since partly refilled, and today's gap sits between 58 and 43.
Same Rumor, Different Rulebooks
Kalshi is not the only venue pricing this, and the other prices are the best argument on the page for reading the rulebook before the number.
| Market and venue | What it pays on | Yes bid / ask, Sept 19 | Sept 11 |
|---|---|---|---|
| Kalshi: OpenAI announces another solution before Jan 1, 2027 | OpenAI, and only OpenAI, announces a complete solution | 62¢ / 66¢ | 84¢ / 85¢ |
| Polymarket: OpenAI announces another solution by Dec 31, 2026 | OpenAI only, under Polymarket's rulebook | 58¢ / 59¢ | not tracked |
| Polymarket: an AI lab announces another solution by Dec 31, 2026 | Any of 40-plus named labs, OpenAI and Anthropic included | 61¢ / 62¢ | 67¢ / 70¢ |
| Polymarket: an AI lab announces another solution by Sept 30, 2026 | Same, with the deadline eleven days out | 17¢ / 19¢ | 43¢ / 44¢ |
| Polymarket: OpenAI announces another solution by Sept 30, 2026 | OpenAI only, same deadline | 17¢ / 18¢ | not tracked |
| Polymarket: an AI lab announces another solution by Dec 31, 2027 | Same, with the far contract's deadline | 83¢ / 84¢ | 87¢ / 89¢ |
| Kalshi: OpenAI announces another solution before Jan 1, 2028 | OpenAI only, same deadline | 76¢ / 84¢ | 91¢ / 92¢ |
| Polymarket: OpenAI announces another solution by Dec 31, 2027 | OpenAI only, same deadline | 82¢ / 83¢ | not tracked |
| Polymarket: Hodge is the next problem an AI lab solves, by Dec 31, 2027 | First qualifying announcement names Hodge | 59¢ / 61¢ | 43¢ / 45¢ |
| Kalshi: Clay awards a Millennium Prize before 2029 | The Clay Institute officially announces a prize award | 16¢ / 22¢ | 27¢ / 32¢ |
Kalshi prices as of 2:07 a.m. ET September 19; Polymarket best bid and ask between 2:06 and 2:12 a.m. ET. Polymarket's OpenAI-only contracts listed on September 9 and were not on this page's September 11 pull. Polymarket's contracts exclude Navier-Stokes announcements and partial results in language nearly identical to Kalshi's.
The row to stare at is the third one, and what it says now is that last week's warning aged well. Every world in which the Kalshi contract settles Yes is a world in which Polymarket's any-lab contract settles Yes, and the Polymarket contract also pays if Anthropic, Google or anyone else on its list gets there first, so the broader contract should trade at or above the narrower one. On September 11 it traded 15 to 17 cents below it. On September 19 the any-lab contract is 61 to 62 and Kalshi's OpenAI-only contract is 62 to 66: the gap is inside Kalshi's four-cent spread, and Polymarket's own OpenAI-only contract at 58 to 59 sits under its any-lab contract, which is the order the rulebooks require. The far pair fixed itself the same way, 83 to 84 for any lab through 2027 against 76 to 84 on Kalshi and 82 to 83 on Polymarket for OpenAI alone. The inversion closed because Kalshi came down 20 cents to meet the crowd on the broader book, not because the broader book came up. That is what one Thursday's mood looks like a week later.
The September 30 rows are the crowd's read on "we are working through how to share these results thoughtfully," and that read fell by more than half in a week. Any lab announcing by September 30 went from 43 to 44 cents to 17 to 19, and OpenAI alone by that date sits at 17 to 18, eleven days out. It fell during the week the Hodge rumor got its strongest write-up yet. The market read the second half of Thursday's report, the part about the announcement taking longer, at least as hard as the first half.
The Hodge row is what the rumor is worth on a market that lets you name the problem. Polymarket's "which problem next" ladder had Hodge at 43 to 45 cents on September 11, up 18 points that day, with Birch and Swinnerton-Dyer at 26 to 27. On September 19 Hodge is 59 to 61 and Birch and Swinnerton-Dyer is 15 to 17: the rumor's first half gained while its second half, the Anthropic-and-Birch-Swinnerton-Dyer half, lost ground. Polymarket's 2026-only ladder, a thinner book with about $4,800 traded on its Hodge row against about $53,500 on the which-problem ladder, has Hodge solved by an AI this year at 53 to 57. That row is not nested inside Kalshi's 62 to 66: the Kalshi contract is OpenAI only, across five problems, and the Polymarket row is Hodge only, across every lab, so neither contains the other and the gap is a comparison, not a contradiction. What the comparison says is modest and consistent with everything else on this page: the crowd that can name the problem puts Hodge by any lab this year a few cents under where Kalshi puts OpenAI announcing anything at all.
The Sentence In OpenAI's Own Post That Matters
OpenAI's account of the Navier-Stokes result contains the line that bears on this contract more than any mathematician's opinion does. The effort began September 1, and the company writes:
"We launched an effort to evaluate it on all open Millennium Prize problems and a few other high-impact problems."
All of them, in parallel groups of agents. Then a smaller group cracked a related Euler result in roughly 50 hours and OpenAI made a choice: "we shifted agents away from the other Millennium Problems." The group that finished ran on the order of 10,000 concurrent agents and landed the result 88 hours after launch, and GPT-6 Astra spent a further 17 hours formalizing it in Lean.
You can read that two ways, and the 34-to-39 spread on Wednesday was roughly the argument between them. One: the other five problems had agents on them for days without a result worth chasing, and OpenAI pulled resources toward the one where a disproof was within reach. Two: OpenAI now has a working template, a model that improved during the run, and a very public reason to spend the "several million dollars" Sébastien Bubeck put on the first run doing it again before the story cools.
The September 10 statement lands on the second reading without settling it. "Substantial progress on another Millennium Prize problem" says the agents went back, or never fully left, and that the company was willing to say so on the record. It does not say complete, does not say which problem, and does not say when. "Working through how to share these results thoughtfully" is the sentence of a company deciding when and how to announce, which is the only event this contract pays on, and a company a week into a priority dispute has reasons to take its time. A dramatic step toward Hodge, the kind that would dominate a news cycle and that "substantial progress" describes exactly, would leave this contract exactly where it is.
That dispute, in short: NYU's Tristan Buckmaster and Anthropic's Levent Alpöge had been working a closely related fluid problem, Buckmaster asked whether OpenAI's model had access to their Codex sessions, and Bubeck answered that "we did not use their prompts or proofs to prompt our models or direct our agents." It is also the live version of the naming trap. Any outside team could announce a complete solution to a qualifying problem and this market would still settle No unless OpenAI's own announcement expressly named them as collaborators. The rumor's second half, that Anthropic may be close on Birch and Swinnerton-Dyer, is that trap in the present tense: an Anthropic announcement settles Polymarket's any-lab contract Yes and leaves Kalshi's exactly where it is.
What 66 Cents Is Actually Buying
In payout terms, the 66-cent ask turns $100 into about $152 if OpenAI makes a qualifying announcement before New Year's, and into nothing if it does not, before Kalshi's fee touches the win. The 84-cent ask on the before-2028 contract turns $100 into about $119. A week ago the near contract's 85-cent ask paid about $118, and on Wednesday night, September 9, its 39-cent ask paid about $256.
So a buyer at 66 is paying 66 cents for OpenAI to finish, decide to announce, and announce within 15 weeks, on a problem it has still not named, in a week when the one report on the subject said the announcement is the part that could take longer, against a Polymarket crowd at 61 to 62 for any lab at all and 58 to 59 for OpenAI alone in the same window. The price is the crowd's probability of the contract's event, and the contract's event is an announcement, complete and named. The rumor is not that event. The Times statement is not that event, and the rulebook says so in its own words.
One last number. The before-2028 contract's first buyer paid 98 cents for "again" on Tuesday evening, half an hour after listing, and by Wednesday morning the price had been argued down to 50. By Friday, September 11, it was 91 bid on a rumor and a sentence; a week later it is 76 bid, 84 asked, and it has not printed a trade since a fraction of a contract at 82 on September 16. That buyer looks less foolish than they did on September 9 and less right than they did on September 11. They also still have not been paid. This board joins the daily Kalshi hub from here, where every price is quoted bid and ask side by side.
Kalshi prices, order-book sizes and the full trade tape were fetched September 19, 2026 at 2:07 a.m. ET; Polymarket prices between 2:06 and 2:12 a.m. ET. The September 11 figures are this page's prior pull at 6:02 a.m. ET that morning. Every claim about the result, the rumor, and OpenAI's statements links to the primary source or the earliest relay we could time. Nothing here is financial advice.
More on this: OpenAI AGI Odds: Altman Moved His Date. The Market Didn't. · OpenAI IPO: The CFO Named The Year. The Market Named Spring · Barcelona Vs Feyenoord Odds: Does The Favorite Survive The Fee? · The Longest Shot On This Board Might Be The Best-Kept Secret · 2026 Governor Races: Every Kalshi Board We Track
