Best AI Model 2026: Traders Called September, Not December

The Quick Answer

Kalshi's best AI board now has two answers, one per contract. The September market checks the Arena text leaderboard on September 30, 2026, and as of 11:52 p.m. ET on September 11 it quotes Claude at 93¢ bid, 94¢ ask. That is up from 85¢ a day earlier. ChatGPT is 1¢ bid, 2¢ ask on the same contract, after trading at 12¢ the day before. The year-end market checks the same leaderboard on December 31. It quotes Claude at a 69¢ mid, up from 62¢ at this page's September 6 pull, and ChatGPT at a 15¢ mid, down from 25¢.

The thing that moved both boards was not a model launch. It was a leaderboard refresh: on September 11 Arena added GPT-6 Astra to the text leaderboard that settles these contracts, and it landed 61st on the view the year-end rules name. The price-blind panel's frozen September 2 numbers, Claude 46% and ChatGPT 15%, stay on the record as written. What the last six cents on the September contract are paying for, how a two-cent ChatGPT squares with a billion weekly users, and where the year-end board still leaves room to argue, are below.

Kalshi's September Board Already Has A Winner

Kalshi runs the best-AI question twice: a monthly contract that checks the leaderboard on the last day of the month and a year-end contract that checks it on December 31. The September contract opened for trading on August 29, so it has been open for 13 days. It does not carry Baidu's Ernie, and it lists Meta's rung as "Muse" rather than "Muse Spark," but otherwise it is the year-end board with a shorter clock. Prices are Kalshi's bid and ask as of 11:52 p.m. ET on September 11; the "day before" column is Kalshi's previous-price reference, the number its 24-hour change is measured from.

Outcome (September 30 Check)	Bid / ask, Sept 11	Day before	Contracts, last 24h	Open interest
Claude (Anthropic)	93¢ / 94¢	85¢	13,348	97,478
ChatGPT (OpenAI)	1¢ / 2¢	12¢	29,168	106,645
Muse (Meta)	2¢ / 3¢	2¢	13,995	29,333
Grok (xAI)	1¢ / 2¢	3¢	1,711	23,662
Gemini (Google)	1¢ / 2¢	2¢	429	73,627
Qwen (Alibaba)	0¢ / 1¢	1¢	0	1,052
Kimi (Moonshot)	0¢ / 1¢	1¢	0	116

Kalshi reports volume and open interest in dollars of contract face value, one dollar per contract, and every contract count on this page uses that unit. The seven asks sum to 105¢ and the seven bids to 98¢; every rung is quoted a cent wide, and Qwen and Kimi have no bid behind them at all.

The row I keep coming back to is ChatGPT, because its 29,168 contracts were the busiest on the board and 23,932 of them, about four-fifths, changed hands after 5 p.m. The trade log puts the whole repricing inside seven minutes. At 5:18:57 p.m. ET buyers started lifting Claude's September offers at 86; at 5:19:13 one seller put 2,500 ChatGPT contracts through the bids in a single second, from 10 down to 6; at 5:23:30 a single buyer took 3,623 Claude contracts at 87 in one second; and between 5:25:55 and 5:26:00 buyers walked Claude's book from 89 to 93 on 4,773 contracts. From 5:18:57 to 5:26:00 the Claude contract traded 11,909 contracts, and 11,897 of them were buyers. Before 5 p.m. it had traded 1,179 all day.

ChatGPT's September contract did not go straight to 2. After the 5:19 sweep it found buyers: 3,744 contracts lifted offers at 5 between 5:48:05 and 5:48:36 p.m. Then a seller hit 2,156 contracts down to 2 at 5:52:33, and at 9:12:21 p.m. another seller put 5,804 contracts through the book in one second at 2 and 1. Buyers took 5,649 contracts back at 2 over the next hour, and the day closed there. On the day the contract traded 13,398 contracts of buying against 15,770 of selling, from an 8 open to a 2 close, and the book at midnight showed 3,769 contracts bid at 1 and 7,556 offered at 2.

The year-end board moved in the same minutes. Between 5:19:53 and 5:23:53 p.m. buyers took 9,985 Claude year-end contracts from 63.4 to 68, and at 5:22:11 one seller put 1,554 ChatGPT year-end contracts through the bids from 19.8 to 18. Two contracts on two clocks, priced by the same buyers inside the same five minutes, which is the first clue about what they were trading on.

The board's other buyer that afternoon came earlier and bought the cheapest thing on it. At 4:45:31 p.m., 33 minutes before the Claude sweep, a single order lifted 13,879 Muse contracts at 2¢ in one second, nearly half of the rung's 29,333 open interest, for roughly $278 at risk against $13,879 of face value if Meta somehow takes the top of the leaderboard on September 30. That one order is most of the reason the rung looks new on a volume screen; it has been listed since August 29, and 13,995 of its 29,404 contracts of lifetime volume printed in the last 24 hours. Whatever that buyer saw, Arena's changelog offers one candidate: Meta's Muse Spark 1.3 (Max) was added to the Agent Arena leaderboard on September 11 and to the Code Arena leaderboard on September 8, and it has no row on the text leaderboard yet.

What Arena Posted On September 11

Every contract on both boards pays on the same source. Kalshi's year-end rules say the check is made on the Arena text leaderboard with the "Remove Style Control" toggle applied, the model with the top Rank (UB) wins, and ties under Rank (UB) break on Arena score, then votes, then the earlier release date; the September contract carries the same tie-break language. Rank (UB) is the leaderboard's own column, and it is shared: a model's Rank (UB) is one plus the number of models the site can say are better than it with statistical confidence, so several models can hold Rank (UB) 1 at once.

That matters here because Arena's changelog for September 11 carries the line "[gpt-6-astra-max] has been added to the [Text Arena] leaderboard," and the leaderboard this page fetched at 11:52 p.m. ET carries a vote cutoff of 1:00 p.m. ET on September 11. The Claude sweep started four hours and 19 minutes after that cutoff. Here is where each lab's best model sits on the settlement view as of that fetch, with 401 models and 8,126,041 votes on the table. This page's September 6 edition showed the leaderboard's default view; the table below is the view the rules name, which orders the same models differently, and its Rank (UB) column is the one the contract reads.

Lab	Best model on the settlement view	Position by score	Rank (UB)	Arena score	Votes
Anthropic	claude-fable-5.1-max	1	1	1510.5 ± 8.4	5,447
Google	gemini-3.8-flash-high	6	1	1494.4 ± 8.5	5,094
Meta	muse-spark-1.2 (xHigh)	10	3	1489.3 ± 10.5	3,229
Alibaba	qwen3.8-max	14	7	1480.9 ± 5.8	16,263
Z.ai (No Contract)	glm-5.3-max	17	9	1477.5 ± 6.5	10,559
Moonshot	kimi-k3-max	22	14	1471.9 ± 5.2	20,911
OpenAI	gpt-5.5-high	24	17	1470.9 ± 3.8	64,953
Baidu (Year-End Board Only)	ernie-5.1	27	17	1468.1 ± 4.6	37,063
xAI	grok-4.5	41	34	1451.0 ± 4.7	29,617
DeepSeek (No Contract)	deepseek-v4-pro	42	35	1450.8 ± 4.0	54,142

Arena lists xAI's models under the organization name "SpaceXAI." The ± figure is the gap between a model's score and the top of its confidence interval, which narrows as votes arrive.

GPT-6 Astra is not in that table because it is not OpenAI's best entry. The new row, gpt-6-astra-max, sits 61st by score at 1441.8 on 2,059 votes, with a Rank (UB) of 35, on the view that settles the year-end contract. On the leaderboard's default view, the one with style control on, it is 24th. OpenAI's best row on the settlement view is gpt-5.5-high, 24th by score; its best row on the default view is a different model, gpt-5.6-sol-xhigh, 19th there. Eight days after launch, the model the September contract had printed at 48¢ on launch day scored 69 points below the leader on the table that pays.

Anthropic's side of the table is the mirror. Its new row, claude-fable-5.1-max, is first by score at 1510.5 on 5,447 votes, and the five highest scores on the leaderboard are all Anthropic's: Fable 5.1 Max, then claude-opus-5-high at 1505.3 on 41,777 votes, claude-opus-5-max at 1504.1, claude-opus-4-6-high at 1502.9 on 71,992 votes and claude-opus-4-6 at 1497.5 on 75,898 votes. Five models share Rank (UB) 1, four of them Anthropic's (Fable 5.1 Max, Opus 5 High, Opus 5 Max and Opus 4.6 High) and the fifth Google's gemini-3.8-flash-high at 1494.4, and the tie-break is Arena score, which Fable 5.1 Max wins by 16 points over the Google entry. The stack is what the September price is paying for: if the thin-vote Fable 5.1 Max row drifts, an Opus model with eight times the votes is five points behind it.

What The Six Cents On Claude Is Paying For

A 94¢ contract turns $100 into about $106 when it pays, before Kalshi's trading fee, and into nothing when it does not. So the question for anyone looking at that offer is what, specifically, the six cents are insuring against, with the check set for 10 a.m. ET on September 30. There are three things, and the price history prices them together.

The first is a new entry. Arena adds models to the text leaderboard days after they ship, not the day of: Gemini 3.8 Flash was added on September 2, and Astra, released September 3, was added September 11. A model that any lab ships in the next week could have a row before the check, and it would need to arrive with a Rank (UB) of 1 and the highest score in that tie, above 1510.5 as of tonight, on enough votes to hold it. The changelog's most recent text-leaderboard addition is Astra's; Muse Spark 1.3 (Max) has reached the agent and code boards but not the text one.

The second is drift. Fable 5.1 Max holds first on 5,447 votes with a plus-or-minus 8.4 interval, and a thin-vote entry can settle downward as votes accumulate; DeepSeek's seat made exactly that point about Meta on September 2. But the tie-break is score among the models sharing Rank (UB) 1, and the next four scores are Anthropic's, on 41,777, 20,296, 71,992 and 75,898 votes. For the crown to leave Anthropic on drift alone, Google's 1494.4 would have to pass all five, a 16-point move for a model whose own interval is plus or minus 8.5.

The third is the rulebook itself, and the table already shows how much one setting is worth: strip style control and Fable 5.1 Max is first, leave it on and claude-fable-5 is, and the same two views put Fable 5.1 Max fourth and claude-fable-5 seventh respectively. Both leaders are Anthropic's today, so the toggle does not change the payout, but it is a live example of a methodology switch reordering the board, and the year-end rules name one view while the September rules carry only the tie-breaks. That is the part no leaderboard table can price, and it is the reason a market this settled still trades a cent wide.

What the book says about those three risks is that traders are paid to take them. At midnight the September contract had 51,178 contracts offered at 94, the largest resting order this page found on either board, against 1,090 bid at 93 and 2,260 bid at 91. Whoever is offering at 94 is willing to sell that much of a favorite for six cents of upside per contract, and after 12,082 contracts of buying in the evening the price had not gone through it. A wall like that is the market's answer to what the tail is worth: about six cents, and there is a great deal for sale at that price.

A Billion Users And A Two-Cent Contract

The ChatGPT rung looks wrong at first glance, and the first glance is about usage. OpenAI said on August 6 that ChatGPT had reached one billion weekly active users, and no other lab has claimed a number like it. A contract that pays if OpenAI has the top-ranked model on September 30, trading at 2¢, with 7,556 contracts offered at that price and 3,769 bid at 1, reads like the market has decided the most-used model in the world cannot be the best one.

That is not what it says. The contract does not pay on users, revenue, or benchmark scores; it pays on one column of one leaderboard, and that column is built from head-to-head votes by people comparing two anonymous answers on Arena's site. Those voters are a different population from ChatGPT's billion, and the model they vote on is whatever OpenAI submits to the arena, at whatever setting. The rung is a bet on gpt-6-astra-max, or a successor, going from 61st by score on the settlement view, and 24th on the default one, to first by September 30, and the billion users have no vote in it.

The price history shows that traders spent launch week learning this the expensive way. On September 3, the day Astra shipped, the September ChatGPT contract printed as high as 48 at 2:43 p.m. ET, and Claude's September contract printed as low as 68 two minutes later. On September 4 at 1:22:02 p.m., one buyer walked the ChatGPT book from 15 to 46 on 13,763 contracts in a single second. The contract closed that day at 18, closed September 10 at 12, and closed September 11 at 2. In payout terms, a 2¢ contract turns $100 into $5,000 if it pays, which is the price the market puts on an OpenAI text-arena row, Astra's or a successor's, clearing a 69-point gap before the September 30 check.

The year-end rung is a different bet, and it is priced like one. ChatGPT's December contract at 15¢ pays on the same leaderboard 111 days out, and OpenAI ships more than once in 111 days. The two prices on the same lab, 2¢ and 15¢, are the clearest statement on either board about what the September 11 leaderboard settled and what it did not.

Who Is Winning The Year-End Board?

Prices are Kalshi mid-quotes as of 11:52 p.m. ET on September 11, the midpoint of the yes bid and ask rounded to the cent. The September 6 and September 2 columns are this page's two previous pulls, the July 24 column is what the page showed when it was first published, and the two panel columns are frozen: the July eight-seat blend and the six-seat re-score of September 2, explained further down. "n/a" means the rung was not on the board or the page that day.

Outcome (December 31 Check)	Kalshi (Sep 11)	Sep 6	Sep 2	Jul 24	July blend (8 seats)	Sept 2 re-score (6 seats)
Claude (Anthropic)	69¢	62¢	68¢	64¢	24%	46%
ChatGPT (OpenAI)	15¢	25¢	21¢	13¢	28%	15%
Gemini (Google)	6¢	6¢	6¢	12¢	25%	14%
Grok (xAI)	6¢	5¢	7¢	7¢	8%	4%
Muse Spark (Meta)	1¢	1¢	2¢	n/a	4%	10%
Qwen (Alibaba)	<1¢	<1¢	1¢	1¢	3%	2%
Kimi (Moonshot)	<1¢	<1¢	<1¢	n/a	n/a	4%
Ernie (Baidu)	<1¢	<1¢	<1¢	n/a	n/a	1%
DeepSeek (No Contract)	n/a	n/a	n/a	n/a	6%	in "someone else"
Someone Else	n/a	n/a	n/a	n/a	3%	4%

Year-end asks sum to 101.2¢ and bids to 97.6¢ across the eight rungs. Claude carries 1,099,164 contracts of open interest and ChatGPT 1,111,501, the two largest on either board.

The row that changed most is ChatGPT, and the change is the one the September 6 edition said the market had refused to make. That page's headline number was that the year-end contract absorbed 274,097 contracts in launch week and moved four cents. Since the September 6 pull it has traded another 92,648 contracts, more of them buyers than sellers, 56,915 to 35,733, and it has fallen ten cents anyway, from 25 to 15, because the sellers chose their moments: 18,434 contracts sold on September 5, a 12.0 print on September 8, and the 1,554-contract sweep at 5:22 p.m. on September 11 after the leaderboard posted. Buyers at 15 and 16 stepped in that evening, 1,098 contracts in the 7 p.m. hour alone, and held it there.

Claude's year-end contract did the opposite, and the interesting part is not the sweep but what came after it. The 5 p.m. hour took it from 62.4 to 68.7 on 10,915 contracts, 10,714 of them buyers. Then, between 7:39 and 9:41 p.m., sellers put 8,912 contracts through the bids at 68.5 and 68.6, and the price did not move. A book that absorbs nearly 9,000 contracts of selling within a tenth of a cent is the difference between the year-end board and the September one: the September contract is a wall of offers at 94 waiting for buyers, and the year-end contract is a two-sided market at 69 with 2,709 contracts bid at 68.5 and 1,109 at 68.6 underneath.

Grok's 6¢ mid, on a 5.9 bid and 6.8 ask, now sits level with Gemini's 6¢ on a 6.2 bid and 6.5 ask, for a lab whose best entry on the settlement view is 41st, and the September rung on the same lab trades 1 bid, 2 ask. Muse Spark's year-end rung slipped to a 1.1 bid and 1.2 ask on 7,839 contracts in the last day, even as its September sibling took the 13,879-contract buyer.

The coding board, which pays on LiveBench's coding column rather than on Arena and has its own page, moved the same direction this week after moving the opposite one in launch week. Anthropic is 66¢ bid, 71¢ ask there, up from 61 and 66 on September 6, and OpenAI is 23¢ bid, 27¢ ask, down from 28 and 33; that board traded 521 Anthropic contracts and 147 OpenAI contracts in the last 24 hours against 13,348 and 29,168 on the September board, so it is still quoted here as a spread rather than a mid. Last week the two leaderboards read one release in opposite directions. This week they read it the same way.

Where The Panel Splits From The Market

Six seats re-scored the race on September 2, the day before Astra shipped: Claude Fable, Claude Opus, Claude Sonnet, GLM 5.2, Kimi K3 and DeepSeek V4. The ChatGPT and Gemini seats were not installed on the machine that produced the run. Every seat saw the same card, the September 2 leaderboard, dated release facts and the settlement rules, and no prices. The seats have not been re-run since, and their numbers are shown as written so that the two leaderboard refreshes since can be graded against them.

Outcome	Claude Fable	Claude Opus	Claude Sonnet	GLM 5.2	Kimi K3	DeepSeek V4
Claude (Anthropic)	40%	57%	55%	45%	42%	38%
ChatGPT (OpenAI)	13%	10%	13%	20%	18%	17%
Gemini (Google)	19%	13%	11%	12%	16%	15%
Muse Spark (Meta)	13%	9%	10%	10%	7%	10%
Grok (xAI)	4%	2%	4%	3%	3%	5%
Kimi (Moonshot)	3%	4%	4%	4%	4%	4%
Qwen (Alibaba)	2.5%	1.5%	1.5%	3%	2%	3%
Ernie (Baidu)	1%	0.5%	0.5%	1%	1%	2%
Someone Else	4.5%	3%	1%	2%	7%	6%

Every seat on this panel is graded against real market settlements — records to date, as the share of graded calls that landed on the right side of settlement: Claude Fable 88% on 3,084 graded calls · Claude Opus 88% on 3,173 graded calls · Claude Sonnet 86% on 3,149 graded calls · GPT 86% on 9,839 graded calls · Gemini 87% on 8,917 graded calls · GLM 84% on 4,104 graded calls · Kimi 86% on 4,478 graded calls · DeepSeek 84% on 4,165 graded calls. Recomputed daily; the full scoreboard is public, and the daily Kalshi hub carries the running record against each day's board.

More live boards: the coding board has Anthropic at 66¢ bid after the same week · Anthropic IPO odds · the any-morning top-AI board that leaves Anthropic off the list.

ChatGPT is where the market has now come all the way back to the panel. In July the six seats had it at 27.3% against a 13¢ price; on September 2 they cut it to 15% against 21¢, and the market spent launch week going the other way to 25¢. Nine days later the year-end contract is at 15¢, level with the blend, and the highest seat, GLM at 20%, is now the one above the price rather than below it. The seats' reason was the leaderboard: GPT-5.6 sat 15th on the September 2 card they scored against, so OpenAI needed a new flagship that cleared roughly 1508 and banked votes before the snapshot. The flagship shipped, the row landed, and it scored 1441.8. The panel did not know that; it priced the odds of it.

Claude is where the gap has widened, from 16 points on September 6 to 23 now, with the market at 69¢ and the panel at 46%. Three of the six cited the Anthropic stack at the top of the board by name, and every seat discounted the market price for the calendar: 120 days on their clock, 111 as of this refresh, is enough for two or three frontier releases, and a December 31 snapshot rewards whoever ships last. One of those releases has now been scored, and the stack is four rows deep on the settlement view. The seats' discount is a discount on the calendar, not on Anthropic, and the calendar is what the September contract no longer has.

Meta is the page's cleanest reversal, and it now runs both ways. In July Claude Opus wrote "Meta is effectively out after repeated reorgs." On the September 2 card Muse Spark 1.2 held fourth on the default view and the panel put Meta near 10%; on the settlement view tonight it is tenth by score and third by Rank (UB), on 3,229 votes with the widest interval in the top ten. Kalshi has the year-end rung at 1.1 bid, 1.2 ask, so the panel's 10% is still roughly nine times the price, the widest premium on the board above a cent. The seats agreed on the number, 7% to 13%, and split on the reason: Claude Fable read the Muse Spark 1.1-to-1.2 version step as a real trajectory, while Claude Opus and DeepSeek priced it as a thin-vote entry that fades as votes arrive. Muse Spark 1.3 (Max) landing on the agent and code boards this week, with no text row, is the next test of that split.

Gemini is the quieter version of the same argument. At 6¢ the market has Google level with Grok, while the panel keeps it at 14%, and the settlement view gives Google something the default view did not: gemini-3.8-flash-high is the only non-Anthropic model sharing Rank (UB) 1, on 5,094 votes with a plus-or-minus 8.5 interval. Claude Fable, the panel's high seat on Google at 19%, read a Flash tier leading Google's board as evidence of an unshipped pro or ultra tier, and Claude Opus named a Q4 Gemini 4 Pro or Ultra as the single most board-moving event on its list, though Opus itself only got to 13%. That is the case for 14% against 6¢. Grok runs the other way, a 6¢ mid for a lab whose best entry is 41st, which the panel priced at 3.5%, shown as 4% in the table.

On home-team bias, the number that matters has not changed since September 2: the three Claude seats averaged 51% on their own maker and the three seats built elsewhere averaged 42% off the same card, a nine-point gap that is smaller than the spread inside the Claude trio, where Claude Fable's 40% sat below GLM's 45% and Kimi's 42%. Both halves of the panel are under the 69¢ market now. December 31 grades all six.

What Each Seat Said

Each seat's argument came down to this, and each cites the September 2 card it was scored against, which had GPT-5.6 15th, Grok 37th and Muse Spark 1.2 fourth on the default view; by the September 6 fetch of the same table those rows read 17th and fifth, and a new xAI entry, grok-4.20-beta1, had landed 28th.

Claude Fable: Google is the top challenger, because a Flash tier leading Google's board implies an unreleased pro or ultra tier that historically lands 10 to 20 points higher. Meta is the crowd's likely underrate on the 1.1-to-1.2 trajectory. OpenAI is the crowd's likely overrate: GPT-5.6 did not crack the top ten, and the late-August release notes show product work, not frontier training.

Claude Opus: The decisive fact is depth, not the lead, because a single Anthropic model regressing costs nothing when the next Anthropic entry is one point behind. Meta's fourth place is softer than it looks on votes an order of magnitude thinner than its neighbors. The single most board-moving event is Google shipping a Gemini 4 Pro or Ultra tier in Q4 with enough votes to land above 1510.

Claude Sonnet: Three successive Anthropic generations clustered at the top, with tens of thousands of votes each, is a cadence advantage, and the market rewards labs already shipping frequently over ones waiting on a single swing. OpenAI is quiet but has the resources to ship a big jump in four months, so it is not collapsed to near zero.

GLM 5.2: OpenAI landing at rank 15 suggests either a pivot away from Arena-chasing or a pending larger release, and it gets 20% because it has the resources and incentive to ship once more before year-end. Meta gets 10% as a dark horse. The one event that would most change this board is OpenAI announcing GPT-5.7 or GPT-6 in October.

Kimi K3: Arena's number one historically rotates every one to three months, so 120 days is long. OpenAI is the most likely dethroner, but GPT-5.6 landing only 15th is a real negative signal, so it is capped at 18. The crowd will over-rate xAI, Grok hype against a rank-37 reality.

DeepSeek V4: Six of the top eight being Anthropic is a real lead, not one lucky model. Meta's young entry is discounted because new Arena entries often fade as votes grow, and the tie-breaks favor vote depth, which helps Anthropic and hurts low-vote entries. The event that would most change the board is an OpenAI or Google frontier release in late November or December.

Two of those lines have now been graded by the leaderboard rather than the market. Claude Opus's "depth, not the lead" is the exact shape of tonight's settlement view, five Anthropic scores before anyone else's. And Claude Fable's line that OpenAI's release notes showed "product work, not frontier training" was the one the launch appeared to answer on September 3, when the September contract touched 48¢; on September 11 the table answered back with a 61st-place row. The seat that wrote it had OpenAI at 13%.

The Bottom Line

The September 11 leaderboard refresh did what launch week could not: it gave both boards a number instead of a guess. Astra's text-arena row landed 69 points under the leader, the September contract went from 86 to 93 in seven minutes of tape, and the year-end contract took four cents off OpenAI and put five onto Anthropic on the day, ten and seven since this page's last pull. Two contracts, one leaderboard, and the difference between them is now purely the calendar: the September check is on the 30th, and the only model on Arena's changelog that has reached the agent and code boards without a text row is Meta's, while the year-end check is 111 days out and the panel's whole case for a 46% rather than a 69% is that 111 days is long enough for someone to ship.

The panel's 15% on ChatGPT and 46% on Claude were written on September 2 and stay on the record as written; one of them the market has now met, and the other it has moved 23 points away from. December 31 grades them. Until then, the number to watch on this page is still the date on the leaderboard, and the next one after September 11 decides whether the six cents on the September contract were worth selling.

July 24 model estimates generated July 24, 2026, price-blind, by eight seats. September 2 re-score generated September 2, 2026, price-blind, by six of those seats; the ChatGPT and Gemini seats were not run, the seats were not re-run after the September 3 launch or the September 11 leaderboard refresh, and the September blend is the equal-weight mean of the six. Kalshi prices as of 11:52 p.m. ET on September 11, 2026 (03:52 UTC September 12), from Kalshi's public market data, order books and trade tape; leaderboard from the Arena text leaderboard's style-control-removed view as fetched at the same time, vote cutoff 1:00 p.m. ET September 11, 2026; model additions from Arena's leaderboard changelog; ChatGPT's user figure from OpenAI's August 6 statement as reported by Yahoo Tech. Models are frequently wrong; the market price reflects real traders' money.

Related Verdicts

The coding board has its own page, with the LiveBench refresh that scored Astra fourteenth on the column that pays and the order book behind its quote: Best AI Coding Model: GPT-6 Shipped. Traders Sold OpenAI.

Kalshi also runs an AI board that pays if a company reaches number one on any morning before 2027, rather than on a single date. As of that page's pull, it left Anthropic and Google off the list entirely, and the panel's numbers there looked nothing like this page: see why the favorite on that board was ranked 14th.

More on this: Barcelona Vs Feyenoord Odds: Does The Favorite Survive The Fee? · Where Does Bitcoin End 2026? Kalshi's Price-Band Odds · The Longest Shot On This Board Might Be The Best-Kept Secret · 2026 Governor Races: Every Kalshi Board We Track · MAC Championship Odds: Ohio Never Has To Play The Favorite