On August 19, OpenAI's chief financial officer did the thing a private company almost never does about its own stock market debut. She named a year. Sarah Friar told employees at an all-hands that OpenAI "will be a public company in 2027", and could arrive sooner if "our business continues to inflect." She also told the room that "the IPO is not a finish line, it is a milestone, another fundraise."
That is the loudest signal this story has produced since the paperwork went in. It is also not the thing the market pays on. Kalshi, a federally regulated exchange for event contracts, runs ten separate contracts on OpenAI's listing, one per deadline, and every one of them settles on a piece of SEC machinery rather than on anything a CFO says in a meeting. So we ran the whole ladder past eight AI models, showed them the contract terms and a card of dated, sourced facts, withheld every single price, and asked each one to build its own timeline.
The Quick Answer
Kalshi prices an OpenAI IPO being confirmed during 2026 in single digits until December, then opens the first serious window in the winter and spring of 2027. Our eight-model panel, working blind, lands in the same place on the autumn contracts and sits below the market on every deadline from December onward. Among the contracts anyone can actually trade at a fair price, the two sides split hardest on February 2027, and the reason is a filing deadline almost nobody outside a securities law firm has heard of.
Nobody handed the models an accounting calendar. One seat raised that February deadline unprompted in the first round, and four others were citing it back by the end of the second. The full ten-rung board, every seat's number, and the four events that would move all of it are below.
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 this kind of contract? Start with how event contracts actually work, because the pricing convention matters here: a contract trades in cents, and the cents are the market's odds. A contract at 20 cents pays out one dollar if the thing happens, so the market is calling it roughly a one-in-five chance.
The Market
| What It Asks | When will OpenAI IPO? Ten contracts, one per deadline, from September 1, 2026 through June 1, 2027 |
| What Settles It | The first of three things: the SEC declares OpenAI's Form S-1 effective, the IPO is priced, or an exchange assigns the stock a ticker |
| What Does Not Settle It | Shares trading, a press report, a stated intention, or the confidential draft already sitting with the SEC |
| Where | Kalshi, a CFTC-regulated event-contract exchange. 18+, availability varies by state |
| Already Decided | The Before July 1, 2026 and Before August 1, 2026 contracts both settled NO |
| Depth | The busiest open contract has traded about 318,000 contracts in its lifetime and still has roughly 150,000 positions sitting open; the thinnest has traded about 20,000 with 7,000 open |
The settlement language is worth reading in full, because it is the reason this board behaves so strangely. Each rung carries its own date in the text; here is the May 1, 2027 contract: "An IPO is confirmed if 1) the SEC declares the company's Form S-1 effective OR 2) the IPO is priced OR 3) a securities exchange has assigned a ticker to it. As long as any of those events occur, the market will immediately resolve to Yes, even if the company does not start trading until after May 1, 2027." Our colleague Eytan Shander pulled that apart trigger by trigger in what actually counts as an IPO here, and the short version is that this contract pays on paperwork, not on a bell.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $189M traded
- 2027 Pro Football Champion · $69M traded
- 2028 U.S. Presidential Election winner? · $61M traded
- 2028 Republican presidential nominee · $59M traded
- Pro Baseball Champion · $58M traded
- Bitcoin price at the end of 2026 · $32M 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.
The Board
| Confirmed Before | Kalshi bid and ask | AI panel |
|---|---|---|
| Sep 1, 2026 | 0 to 1¢ | 1.0% |
| Oct 1, 2026 | 0 to 2¢ | 2.0% |
| Nov 1, 2026 | 1 to 6¢ | 4.0% |
| Dec 1, 2026 | 10 to 15¢ | 7.5% |
| Jan 1, 2027 | 16 to 21¢ | 12.5% |
| Feb 1, 2027 | 28 to 38¢ | 18.0% |
| Mar 1, 2027 | 29 to 67¢ | 27.5% |
| Apr 1, 2027 | 26 to 75¢ | 38.5% |
| May 1, 2027 | 37 to 75¢ | 49.0% |
| Jun 1, 2027 | 45 to 82¢ | 58.5% |
Prices fetched August 23, 2026. The panel column is the blend we publish on every board: the median of eight independent model estimates after a revision round, not the mean of all eight. Read the two price columns as a pair rather than as a single number. A bid is the best standing offer to buy, an ask the best offer to sell. Two of these ten contracts have no bid at all, September and October, which means nobody is offering to buy them at any price. Four more are quoted tightly enough to read as real prices: November, December, January and February. November got there the hard way, its bid sliding from 5 cents to 1 over Sunday morning while it traded, which leaves a quote wider than the price it brackets. From March 2027 out the gap never closes below 37 cents and reaches 49 on April, and a midpoint drawn between a 29-cent bid and a 67-cent ask is an arithmetic exercise rather than a price anyone is trading.
That distinction does most of the work in this piece, and we will come back to it.
Every Seat's Number
Each model priced all ten rungs in a single pass, from one internal view of when the first trigger fires, rather than being asked about each deadline separately. Then every seat read the other seven curves with the names stripped off and was invited to revise. Four columns are shown here; the shape of each curve is the same story.
| Model | Dec 1, 2026 | Feb 1, 2027 | Apr 1, 2027 | Jun 1, 2027 |
|---|---|---|---|---|
| GPT | 9.5% | 24% | 49% | 70% |
| Gemini | 7% | 18% | 40% | 65% |
| Claude Opus | 5.5% | 13% | 27% | 45% |
| Claude Sonnet | 7% | 17% | 32% | 48% |
| Claude Fable | 7% | 15% | 31% | 48% |
| GLM | 8% | 24% | 43% | 62% |
| Kimi | 8% | 18% | 37% | 55% |
| DeepSeek | 8% | 25% | 55% | 85% |
| Panel Median | 7.5% | 18% | 38.5% | 58.5% |
| Kalshi Bid | 10¢ | 28¢ | 26¢ | 45¢ |
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 82% on 616 graded calls · Claude Opus 83% on 686 graded calls · Claude Sonnet 81% on 664 graded calls · GPT 84% on 5,836 graded calls · Gemini 86% on 4,570 graded calls · Kimi 82% on 2,442 graded calls · GLM 81% on 2,465 graded calls · DeepSeek 80% on 2,490 graded calls. Recomputed daily; the full scoreboard is public.
These are model estimates, not predictions of fact and not financial advice. Verdicts generated August 23, 2026; prices as of August 23, 2026.
Every number on this page gets graded in public once these contracts settle, and the running tally lives on the full graded scoreboard.
More live boards from the same panel: Anthropic's IPO ladder, where a confirmed listing before December 1, 2026 is bid at 89¢ against OpenAI's 10¢ · the 2026 top-AI-model board, where OpenAI is bid at 31¢ to finish the year with the number-one model. Prices fetched August 23, 2026.
Why This Autumn Is Close To Mechanically Impossible
The panel converged on the 2026 rungs faster than on anything else, and the reason is a rule rather than a forecast.
OpenAI submitted a confidential draft registration statement to the SEC on June 8, 2026, a week after Anthropic did the same. TechCrunch reported the filing alongside a company statement that reads very differently now than the CFO's August remarks: "we have not decided on timing yet; it may be a while because there are things we want to do that are likely easier as a private company."
A confidential draft is invisible and settles nothing. To get from there to any of the three triggers, a company has to flip that draft into a public filing, and the SEC's guidance for issuers using the confidential review process requires the public filing to land at least 15 days before the road show, the pitch tour where bankers take the company around to big investors, or at least 15 days before the requested effective date if there is no road show. Effectiveness and pricing then arrive at the end of the road show, not the start.
That makes the public flip an observable, dated starting gun. We ran a full-text search of the SEC's EDGAR system on August 23, 2026 for OpenAI Group PBC and for any S-1 filed by OpenAI. There are hundreds of hits and not one of them is OpenAI. Every result is somebody else: mutual funds disclosing their private stake, and unrelated issuers naming OpenAI as a customer or a competitor. The gun has not been fired.
The starting gun for this entire board is one public filing on EDGAR. As of August 23, 2026, it has not been fired, and nothing can settle before it is.
The models added a road show of one to two weeks on top of those 15 days, which puts the earliest settlement from a public filing made today in late September or October. Nothing in the record points to a flip happening this week. All eight seats put September at or near 1% and October at or near 2% for exactly that reason.
The Mid-February Wall
Here is where the panel earned its keep. Nobody handed the models a calendar of accounting deadlines. Claude Opus raised a constraint no other seat had priced in the first round, and by the end of the second, four of them were citing it back.
Financial statements in a registration statement go stale. Under the SEC's age-of-financial-statements rules, the numbers in the filing cannot be older than 135 days at the moment the registration goes effective. For a company with a December year end, financials as of September 30 hit that wall around February 12. After that the company needs audited full-year figures, and audited figures are required outright once effectiveness falls more than 45 days past the fiscal year end, which is roughly February 14.
So the calendar has a gap in it. A deal that wants to price on September-quarter numbers has to be done by mid-February. A deal that misses that has to wait for the audit, which realistically means late March at the earliest. The window in between is not impossible, but it is narrow and expensive, and that is precisely where Kalshi's February contract sits.
The panel priced February 1 at 18% against a 28-cent bid. That is the widest gap on any contract here you could actually trade at a fair price, and the mechanism behind it is a date on an accountant's calendar.
Where The Panel Changed Its Mind
Seven of the eight seats moved in the revision round, some by three points and some by thirty.
Gemini cut its June 2027 number from 85% to 65% and said why:
"I was persuaded by peers who emphasized the structural friction of a $1T valuation target, massive $600B compute commitments, and SpaceX's recent rocky aftermarket, all of which heavily increase the hazard of this mega-IPO slipping into H2 2027." — Gemini
The aftermarket is how a stock trades in the days after it lists, and SpaceX's is the cautionary tale sitting in front of every large listing this year. More on that below.
GPT made the largest cut on the near deadlines, taking November from 15% to 5.5% and December from 23% to 9.5%:
"[T]he other forecasters sharpened a point I underweighted: the public S-1 flip is not just a two-week precursor, but realistically a 4-6 week countdown once roadshow, SEC comments, and mega-IPO preparation are included." — GPT
DeepSeek made the deepest cut of anyone in the winter, taking January from 40% to 15% and February from 55% to 25%:
"[A] Q1 2027 deal using Q3-2026 financials would go stale by mid-February, forcing either a rushed timeline or a wait for audited FY2026 numbers, which pushes the public S-1 flip and roadshow into March–April." — DeepSeek
Kimi held, and held for a stated reason, which is the version of this exercise that matters most. It was rejecting another seat's much higher January number on the grounds that the number smuggled in too much chance of the trigger firing in November or December:
"...it implies ~25 points of Nov–Dec 2026 hazard, contradicting Friar's on-record 2027 guidance delivered four days before this pricing date..." — Kimi
The one thing all eight seats named as the event that would most change their curve was identical: a public S-1 appearing on EDGAR.
The Court Fight That Cleared The Runway
None of this timeline existed in a usable form a year ago, and the reason is a lawsuit.
Elon Musk sued OpenAI and Sam Altman in 2024, alleging they had broken a promise to run the venture as a nonprofit. He asked the court to return more than $130 billion to OpenAI's nonprofit arm, to remove Altman and Greg Brockman from their roles, and to unwind the corporate restructuring outright. Any one of those outcomes would have made a listing unthinkable on this timeframe.
On May 18, 2026, a federal jury in Oakland threw out every claim in under two hours, finding the claims fell outside a three-year statute of limitations. District Judge Yvonne Gonzalez Rogers adopted the advisory verdict. Musk's lead attorney said they would appeal, and Musk himself called the outcome a technicality.
The restructuring that fight was about had already closed on October 28, 2025. The nonprofit became the OpenAI Foundation and holds roughly 26% of the for-profit OpenAI Group PBC; Microsoft holds about 27%, worth roughly $135 billion at the time; the attorneys general of Delaware and California both issued no-objection statements. That structure is what an underwriter can actually sell. The May verdict is what let anyone believe it would hold.
Strip out the legal history and the board looks arbitrary. Put it back and the sequence is plain: the restructuring closed in October 2025, the jury cleared it in May 2026, and the paperwork followed in June.
Three weeks after the verdict, the confidential draft went to the SEC. That is the order the record shows, and a successful appeal is the one thing in it still capable of putting sand in the gears.
Where The Market And The Panel Actually Disagree
The headline version of this board says the panel is far below the market almost everywhere. On the rungs where that gap looks biggest, it is mostly an illusion created by an empty order book.
Look at March 2027. The panel says 27.5%. The screen shows a bid of 29 cents and an ask of 67 cents, 38 apart. Both are real orders. A seller has to take the 29, the most anyone will pay; a buyer has to take the 67, the least anyone will sell for. The distance between them is larger than any edge the panel is claiming. April is worse, at 26 bid and 75 ask.
Now look at the four tightly quoted contracts. November trades 1 to 6, December 10 to 15, January 16 to 21, February 28 to 38. November's quote is wider than its own midpoint, so the panel's 4% and the market's 3.5 are not really arguing about anything. On the other three they are, and the panel is lower every time: 7.5% against a 12.5-cent midpoint, 12.5% against 18.5, and 18% against 33.
The disagreement is less about whether OpenAI eventually goes public than about how much of 2027 gets used up before the first trigger fires, and the panel keeps putting more of its probability past this ladder's last rung than the market does. Six of the eight seats named the same reason: "public company in 2027" covers twelve months, this ladder only reaches June, and the CFO gave no month.
One more honest note on depth. Kalshi's 24-hour volume field reads zero on half the contracts on this board, and it is unreliable enough that we do not use it. Lifetime volume and open positions, which we do trust, show the September contract alone holding more open positions than all four spring 2027 contracts combined. The market's conviction is concentrated exactly where its conviction is easiest: on the months that are already gone.
What Would Change The Panel's Mind
Four events, each checkable, each with the direction it pushes.
A public S-1 appears on EDGAR. Every seat named this first. It starts a hard countdown of roughly four to six weeks to a settlement trigger and would immediately pull enormous probability forward into whichever rung sits at the end of that countdown. This is the one to watch, and it is free to watch.
Pre-IPO investor meetings get scheduled. As of the last reporting on it, OpenAI had held none and had set no timeline. Those meetings precede a road show by weeks, so their appearance is an earlier and softer version of the same signal. Pushes the curve up.
Audited full-year 2026 financials land, or visibly slip. The mid-February staleness wall is the panel's central mechanism. Audited numbers arriving on schedule keep late March and April live; any sign of an audit dragging pushes mass toward May, June and beyond this ladder entirely.
The valuation target moves. Reporting in June said Altman treated any reduction from the roughly $1 trillion target as a nonstarter after advisers warned a listing might not be met with enough enthusiasm. SpaceX had gone public on June 12, raised more than $85 billion, run to $225 a share and fallen to $153 within two weeks. If OpenAI publicly accepts a lower number, the calendar gets much shorter. If it holds the line, the calendar stretches.
The Settlement Timeline
| Date | What happens |
|---|---|
| Sep 1, 2026 | Deadline passes on the first open contract. EDGAR is empty, and the 15-day rule means a filing made now could not settle it |
| Feb 12-14, 2027 | September-quarter financials go stale; audited full-year numbers become mandatory |
| Jun 1, 2027 | Last rung on this ladder expires |
| Any Day | A public S-1 filing settles nothing by itself, but starts the four-to-six week clock |
We re-score this board whenever the story moves rather than on a schedule. Prices and panel numbers on this page are stamped August 23, 2026.
Anthropic Is The Control Group
The cleanest way to see what this market believes is to put it next to the one that shares its industry, its filing month, and its settlement rules.
Anthropic filed its confidential draft on June 1, 2026, a week before OpenAI. Its Kalshi ladder is bid at 42 cents for a confirmation before October 1 this year and 89 cents before December 1. OpenAI's before-December contract is bid at 10. Both fetched August 23, 2026. Same sector, same paperwork, same rulebook, and traders think one of them is nearly done and the other has not started.
Whether that is right is a separate question, and our panel took Anthropic's ladder apart on August 16 and came in well under the market there too. But it does clarify what the OpenAI board is actually saying. It is not saying the IPO is in doubt. It is saying the paperwork is slower than the headlines.
- Kalshi OpenAI IPO Market: What Actually Counts As An IPO?
- Anthropic IPO Odds: When Will Anthropic Go Public?
- Bitcoin Set Its 2026 High In January And Never Got Close Again
- The East Pacific Has One Major Hurricane. Traders Expect Six.
- Who Wins The 2027 NBA Title? Kalshi Championship Odds
The Bottom Line
A CFO put a year on the record. The contracts do not pay on years, they pay on an SEC stamp, and the paper trail that produces that stamp has not started in public yet. That is why the autumn contracts trade in low single digits, and why our panel, working from the same rulebook and no prices at all, lands in the same place.
The real argument is about the six months after New Year's, and it is smaller than the screen makes it look. On the three contracts with a genuine two-sided quote, the panel sits five to fifteen points under the market, with the widest gap in February and a filing deadline behind it. Everywhere further out, the order book is too wide for the disagreement to mean anything yet.
Watch EDGAR. When OpenAI's S-1 goes public, every number on this page becomes stale within the hour, and we will re-score all ten rungs the same day.
Prediction market contracts are 18+ and offered on CFTC-regulated exchanges; availability varies by state. To be explicit: everything above is a set of model estimates and market observations, not financial advice, and not a recommendation to take a position. Contracts on thin books can be expensive to enter and to exit.
Hero illustration: OddsShopper, in the house collage style. Sam Altman photo by Steve Jurvetson, licensed CC BY 2.0; photos cropped, toned, and composited.
