Updated September 5, 2026 · prices as of August 16, 2026 · by Jake Hari
The Quick Answer
Kalshi says the big launches are on the way. GPT-6 before 2027 is priced at 78¢, and OpenAI's Astra before November at 88.5¢. Our eight-model panel, shown the settlement rules and the evidence but never a price, put those at 18.6% and 41.3%.
The panel landed under the market on eight of the 12 boards, and the reason is not doubt about the technology. Most of these contracts pay on a name, an access tier or a buyer, and all 16 deadlines Kalshi has already graded on these same subjects this year settled No. The market is pricing whether the product ships. The panel is pricing whether the sentence in the contract gets satisfied.
The clearest case the other way is the iPhone 18. Kalshi prices it at 12.5¢ before 2027. The panel says 66%, and the whole argument is whether the Pro model counts.
Prices As Of August 16, 2026
Every price on this page is the midpoint of Kalshi's bid and ask, read on August 16, 2026. Several of the nearer deadlines have passed since then, and none of those settlements are reflected below. Where a row's deadline has gone by, it reads as it was priced that day.
The Board
| Subject | What has to happen | By | Kalshi | AI blend |
|---|---|---|---|---|
| OpenAI GPT-6 | OpenAI releases a model called GPT-6 or higher | Jan 1, 2027 | 78¢ | 18.6% |
| NVIDIA Vera Rubin | Vera Rubin on sale to the US general public | Nov 1, 2026 | 63.5¢ | 5.4% |
| Apple iPhone 18 | Apple releases the iPhone 18 to the public | Jan 1, 2027 | 12.5¢ | 66.0% |
| OpenAI Astra | Astra released publicly, any brand name | Nov 1, 2026 | 88.5¢ | 41.3% |
| xAI Grok 5 | xAI releases Grok 5 publicly | Jan 1, 2027 | 67¢ | 31.8% |
| Tesla Optimus | Optimus or another humanoid robot on public sale | Dec 31, 2026 | 9.5¢ | 41.5% |
| Apple Foldable iPhone | Apple announces a foldable phone | Dec 31, 2026 | 92.5¢ | 65.4% |
| Anthropic Opus 6 | Anthropic releases Opus 6 publicly | Nov 1, 2026 | 12.5¢ | 4.9% |
| Google Gemini 3.5 Pro | Google releases Gemini 3.5 Pro or higher | Aug 31, 2026 | 9.5¢ | 14.0% |
| Meta Llama 5 | Meta releases Llama 5 publicly | Jan 1, 2027 | 6.5¢ | 3.4% |
| Anthropic Mythos | Anthropic releases a model called Mythos publicly | Jan 1, 2027 | 16.5¢ | 14.4% |
| Sony PlayStation 6 | Sony announces the PS6 | Dec 31, 2026 | 7¢ | 8.1% |
Kalshi midpoints as of August 16, 2026, sorted by the size of the disagreement, largest first. The "what has to happen" column compresses the settlement rule; the exact wording behind the biggest gaps is quoted below. The Gemini 3.5 Pro deadline has since passed and that row is shown as it was priced.
Model estimates generated August 16, 2026, price-blind, by eight seats (Claude Fable, Claude Opus, Claude Sonnet, ChatGPT, Gemini, GLM, Kimi and DeepSeek). Each seat priced all 12 subjects in a single pass, then read the other seven boards anonymously and revised; every blend above is the post-revision number. Models are frequently wrong; the market price reflects real traders' money. Every seat is graded against real settlements on the full scoreboard.
How Often Does A Kalshi Release Contract Actually Pay?
Rarely, and that base rate is the most useful number on this page.
We pulled every release and launch contract Kalshi has settled across 27 series covering AI models, phones, chips, consoles, cars, games, albums and shows. Of the 66 that have resolved, 21 paid out, a hit rate under a third.
Narrow it to AI models, where shipping is supposedly fastest, and it gets harsher: 8 of 44 settled Yes. The far rungs are where the optimism dies. Contracts listed with more than 120 days to run have gone 0 for 4, while rungs listed a few weeks out pay most of the time, because a deadline that close usually gets listed when something is visibly imminent. This board's biggest prices sit on the far rungs.
Then there is the local record. Sixteen deadlines have already passed on these 12 subjects during 2026, and every one settled No. Seven were weekly Gemini 3.5 Pro rungs, three were Mythos, and the rest fell on GPT-6, Astra, Grok 5, the iPhone 18 and Vera Rubin. Not one product on this page has cleared a deadline the exchange set for it, and that record is most of why the panel sits under the market.
Will GPT-6 Ship Before 2027?
The GPT-6 contract is the widest gap on the board and the cleanest illustration of what these markets actually price.
The rule pays if OpenAI releases "a model called GPT-6 or greater" before January 1, 2027, publicly and outside a closed beta. A high-cost subscription tier counts. The market says 78¢. The panel says 18.6%.
The gap comes from a document Kalshi wrote into its own contract. On August 1, 2026, OpenAI disclosed a model it calls Astra, describing it as "our next major model" and crediting an internal version with ten long-standing open problems in mathematics and theoretical computer science. Kalshi's Astra terms say a qualifying release includes any model publicly named Astra, or confirmed to share its training run, under any commercial name or access tier. Nowhere do they say Astra will be called GPT-6.
So the market is pricing "OpenAI ships its next frontier model this year," which is a reasonable thing to expect. The panel is pricing the sentence in the contract, which needs the number 6 or higher on the box.
"Astra is the obvious next ship, but the contract needs the GPT-6 name and OpenAI has not committed to it. A disclosure is not a release, and a differently-branded release does not pay this rung." — Kimi
The Astra board splits the same way for a different reason. Kalshi prices Astra publicly out before November 1 at 88.5¢. The panel says 41.3%. Every seat took the disclosure seriously, and every seat noticed it came with no date, no pricing, no model card and no availability. The panel's blend does not cross 50% until the December rung, where the market is already at 96¢.
For the full GPT-2 through GPT-5 cadence against this ladder, see our GPT-6 release date odds board.
Does An iPhone 18 Pro Count As An iPhone 18?
This is the board where the panel sits furthest above the market, and it is the most-traded contract on the page, with 403,401 contracts of open interest.
The rule is short: Apple has to release the "iPhone 18" to the public before October 1, 2026. Kalshi prices that at 6.5¢, and 12.5¢ for the same question by the start of 2027. The panel says 53.8% and 66.0%.
Both sides agree on the facts. On August 12, 2026, Apple manufacturer Pegatron confirmed a split launch at its earnings conference: the iPhone 18 Pro lineup is still planned for September 2026, while the base iPhone 18 may not appear until the first quarter of 2027. Nobody on either side thinks the base model ships in the fall. The whole disagreement is whether "iPhone 18 Pro" satisfies "iPhone 18."
"Pegatron confirms the split. This is a coin-flip on whether the grader reads iPhone 18 Pro as iPhone 18, not on Apple's factory." — Claude Opus
Six of the eight seats landed between 55% and 68% on that reading. The market, at 6.5¢, is close to a flat statement that only the base model counts. The best guide we have is how this exchange has actually graded: three Mythos rungs settled No because a restricted release did not clear a public-release clause. That strict reading favors the market here, and it deserves more weight than the panel's confidence.
Is Vera Rubin "Released" If Only Hyperscalers Can Buy It?
NVIDIA said on May 31, 2026 that Vera Rubin had ramped into full production, with production shipments set to begin this fall. The market read that and priced a November 1 release at 63.5¢ and a January one at 85.5¢.
The panel read the settlement rule instead. It requires the product to be "available for purchase, subscription, or download by the general public" through official channels or authorized retailers in the United States, and not labeled beta, early access or preview. Rack-scale datacenter silicon shipping to hyperscalers does not obviously clear that bar, and the exchange already graded the August rung No.
"Fall shipments are hyperscaler-channel; rules demand general-public US purchase. The tail is the resolver accepting OEM or distributor orderability as available for purchase." — Claude Fable
The blend is 5.4% on the November rung against 63.5¢, the second-widest gap on the page. It is also the thinnest market here, with 158 contracts of open interest on that rung and none on the January rung, so neither number is standing on much.
Does Optimus Pay If Someone Else's Robot Goes On Sale?
The Optimus contract produced the widest split of the run, from 10% to 88% across eight seats, and it comes down to a parenthetical. The rule pays if "the Tesla Optimus (or another humanoid robot)" is on sale to the general public before December 31, 2026.
Read narrowly, this is a bet on Tesla, and Tesla is not close. As of April 2026, first-generation production lines were being installed at Fremont, a Texas line was designed for volume starting in 2027, and no consumer sale had been announced. Read literally, the contract pays if any humanoid robot is on public sale, which several seats argued is already true.
"The verbatim rule pays on 'another humanoid robot' on sale to the general public, which consumer-purchasable humanoids already satisfy; only resolver strictness is in question." — Claude Fable
The seats citing specific third-party robots were reasoning from their own knowledge, not from the data card. Unitree's own G1 product page lists a price from $13.5K but shows no purchase button, and the education configuration is quote-only through sales contact. Whether that clears "on sale to the general public" is exactly the ambiguity the panel is describing, and the exchange has not graded it.
The market sits at 9.5¢, which is the crowd saying the narrow Tesla reading will be applied. The panel's 41.5% is not a forecast about robotics. It is eight models split on a resolver's future decision, and a blend that averages a 10 and an 88 is describing a coin flip, not a number to lean on.
What About The Quieter Boards?
Anthropic Mythos, 16.5¢, panel 14.4%. The rare row where the two sides nearly agree, and both are pricing a naming decision rather than a model. Mythos was announced April 7, 2026. On June 9, Anthropic shipped Mythos 5, restricted to approved organizations, alongside Fable 5, the safeguarded public sibling. Three Mythos rungs have settled No since, which tells you the exchange does not count a restricted-access ship. Claude Opus, the most bullish seat at 25%, argued the contract "pays only if Anthropic attaches Mythos to a paid public tier," a policy reversal rather than an engineering milestone. The full Mythos board has the settlement history.
Apple Foldable iPhone, 92.5¢, panel 65.4%. The bar is an announcement, not a launch, and Apple's September event is a hard date. The market is nearly certain and the panel only mostly so, because the case rests on a supply-chain leak chain.
"Announcement-only bar; leaker reports say on schedule with yields solved; Apple's September event is a hard date. The risk is the leak chain being one stale rumor." — Claude Fable
Meta Llama 5, 6.5¢, panel 3.4%. In April 2026 Meta Superintelligence Labs released Muse Spark as a replacement for Llama, and the encyclopedia record now writes the Llama family in the past tense. The contract pays only on a model named Llama 5. Several low-quality sites claim one already shipped in April. The still-open contract says otherwise.
Anthropic Opus 6, 12.5¢, panel 4.9%. The row where the market has the better-looking case. Anthropic's 2026 cadence ran from Opus 4.6 in February to Opus 5 in late July, a major release every six to ten weeks. The panel weighted the naming step from 5 to 6 more heavily than the calendar. This is a disagreement to watch rather than one to lean on.
Google Gemini 3.5 Pro, 9.5¢, panel 14.0%. The only near-term board, and the one with the ugliest record: seven consecutive weekly rungs settled No between June 30 and August 7. On July 21, Google released three Gemini models and none of them was 3.5 Pro, with Bloomberg reporting internal delays against performance goals. The panel sat modestly above the market because the contract pays on "3.5 Pro or greater" and Google ships often. This rung's August 31 deadline has since passed, and its settlement is not reflected here.
Sony PlayStation 6, 7¢, panel 8.1%. The tightest agreement on the board. The bar is an announcement, and the panel still lands in single digits: no primary Sony announcement of a successor was located. Sony named the PS5 in April 2019 and shipped it 19 months later, so an announcement inside 2026 would point at a 2028 console rather than a 2027 one.
xAI Grok 5, 67¢, panel 31.8%. The ladder barely moves until its final rung and then leaps, a bet that the version number eventually jumps. xAI's own developer release notes list Grok 4.6 in August as the newest model, and Grok 5 appears nowhere on the page.
Where Did The Seats Disagree With Each Other?
Every seat priced the board alone, then read the other seven anonymously and could revise. Three arguments moved real probability.
The Optimus parenthetical was the big one. Seven of eight seats raised their number after another forecaster pointed at those four words, and the blend went from 17.8% to 41.5%.
"Tesla consumer sale remains unlikely, but the rule covers another humanoid robot, making the contract broader than Tesla's roadmap." — ChatGPT
The Mythos argument ran the other way and did more damage. DeepSeek came in at 45%, treating the June restricted release as the first step toward a public one, and left at 5%.
"Three settled NOs prove restricted Mythos fails; public sibling is Fable. Naming and access tier make a public Mythos release very unlikely." — DeepSeek
Kimi made the same move, from 55% to 22%. The blend dropped from 25.8% to 14.4%, about two points from the market.
The one reversal that should make you cautious is DeepSeek on the iPhone 18, which went from 5% to 90% after reading the others. A swing that size on a settlement question is a seat discovering an argument, not gathering evidence. It is a good reason to read that row's blend as the wide range it is.
The Bottom Line
Stop reading these as questions about engineering. Almost every disagreement on this page is a disagreement about a sentence. GPT-6 needs a number on a box. Mythos needs a name attached to a public tier, and Anthropic deliberately gave the public tier a different name. Vera Rubin needs a general-public buyer for a product built for hyperscalers. Optimus needs a resolver to decide what a parenthetical means. The iPhone 18 needs someone to rule on whether Pro counts.
Only Gemini 3.5 Pro and the PlayStation 6 are plain questions about whether a company finishes something on time, and those are the two rows where the market and the panel agree most closely. On Opus 6 and Grok 5 the panel's pessimism is a judgment about company cadence, and reasonable people can take the other side of both. On the rest it is closer to arithmetic on a rulebook.
Every number here will be graded in public when these markets settle.
Prices on this page are Kalshi's book. If you also trade on Polymarket, code OS4 gets new users a $50 trading bonus on a $10 deposit — affiliate link; terms as stated by Polymarket; 18+, availability varies by state.
More on this: GPT-6 Release Date Odds: The Name OpenAI Still Won't Say · GTA 6 PC Release Date: The Pattern Rockstar Never Breaks · Elder Scrolls 6 Release Date Odds: When Does It Actually Drop?
