Anthropic Mythos Release Date Odds: Kalshi Prices 16-17% Before 2027

Kalshi's "When will Anthropic release Mythos?" event has traded more than 1.9 million contracts across its life, over half of them on strikes that have already settled, and the board's read is blunt: a public Mythos release before September 1 trades at 2 cents, before December 1 last traded at 16 cents, and the year-end strike (before January 1, 2027) sits at 16 to 17 cents. In other words, the market prices about an 83% chance that no public Mythos ships in 2026 at all. Three earlier strikes have already settled No, one after another, and that streak is the real story: release-date markets look like tech-hype markets, but they settle like rules markets. The fine print that decided those three strikes is doing most of the work on the live ones too, and we will get to exactly what it says.

The Quick Answer

The Kalshi Mythos board prices a public Anthropic Mythos release at 1-2% before September 1, 4-6% before October 1, 7-10% before November 1, 11-16% before December 1, and 16-17% before January 1, 2027. The Jun 15, Jul 1, and Aug 1 strikes all settled No because settlement requires two things at once: a model actually named Mythos, and a release "to the public, outside of a closed beta." Anthropic shipped the same underlying model to everyone on June 9 as Claude Fable 5, and the June strike still settled No. The full strike ladder, the fee math, and the clause that decides the year-end strike are below.

The Board

Live prices from Kalshi's order book as of August 12, 2026. A Yes price in cents reads directly as an implied probability: a 17-cent Yes is a 17% chance, before fees.

The row I keep coming back to is December against January. The year-end strike has the tightest book on the board in proportional terms, one cent wide on a 16.5-cent midpoint (about a 6% spread, versus September's one-cent spread on a 1.5-cent midpoint, closer to 67%), while December 1 is quoted five cents wide at 11/16. Using midpoints, the market adds only about three points of probability for the entire month of December, so the crowd is not pricing a holiday-window launch event. And the midpoint read is not executable: to actually isolate December you would buy the January Yes at the 17-cent ask and sell the December Yes at the 11-cent bid, a 6-cent debit for a month the mids price at 3. The book charges double the midpoint read to own that window, which is what a thin board looks like when you try to trade its shape instead of admiring it. It is pricing a thin, almost flat monthly trickle: a few points for September, a few for October, a few for November. No single month is the story, and that flat shape is what you get from traders who think the binding constraint is not engineering cadence but the definition of "release" itself.

Free: The Weekly PM Market Brief — the week's biggest market-vs-model gaps across Kalshi's boards and what's spiking next. One email, Sundays. Sign up in the box at the end of this article.

The Fine Print That Already Settled Three Strikes

Every strike on this board asks two questions at once, and both have to be Yes for the contract to pay. First, the name: the market resolves on Anthropic releasing "a model called Mythos," not a Mythos-class model wearing a different label. Second, the access: the release must be "to the public, outside of a closed beta." The market summary allows "a high-cost subscription tier," and the formal contract terms set the bar even lower, counting any model "released to only some paid members." The two clauses interact, and the interaction is the whole game: an approved-organizations program is a gated program in the settlement's eyes no matter what members pay, so the paid-members carve-out only rescues a tier the public can actually sign up for. The settlements applied exactly that reading, with approved organizations already running Mythos 5 while strike after strike paid No.

June 2026 stress-tested both conditions inside three weeks. On June 9, Anthropic released Claude Fable 5 to the public and launched Claude Mythos 5 alongside it, the same underlying model, offered only to approved organizations. On June 12 a federal export-control directive ordered access suspended, and within a day both models were offline worldwide; the controls were lifted June 30, the public model came back, and Mythos 5 went right back behind the approved-organizations gate. Through all of it, the June 15 strike settled No, then July 1, then August 1. Anthropic did ship a model named Mythos, but it stayed restricted; the model the public could use was named Fable. Under the rules, there has never been a public model named Mythos, and Kalshi's settlements confirmed that reading three times.

That is the trap in the release-date class, and it is not unique to AI: fan-heavy boards reward reading the settlement definition before the headline, whether the unknown is the winner, as in the Oscars Best Picture race, or the date, as here. Casual money reads "Mythos announced" headlines and buys Yes; the contract settles on the name and the access. Think of it as the prediction-market version of a box-score prop: the performance can be real while the stat line that actually settles your ticket stays empty. If you have read our breakdown of how Kalshi boards price player props against DFS apps, that discipline carries over directly. The market pays the people who read the scoring rules before the lineup lock, and here there are exactly two of them.

The Base Rate Under The Price

Stokastic readers think in projections, so treat the launch cadence as the prior and the market price as the update. Anthropic's Claude 5 generation is barely two months old: Claude Fable 5 launched it on June 9, 2026, roughly a year after the Claude 4 generation arrived, with point releases landing every few months between generations. If this market asked "will Anthropic ship a meaningful new model within six months," that cadence alone would justify a much fatter Yes than anything on this ladder.

But the market is asking whether one specific name crosses one specific access threshold by a date, and that adds three discount factors the raw cadence prior never sees. Naming risk: the broadly released product can keep carrying the Fable label while Mythos stays restricted, and this board never pays. Packaging risk: Anthropic simply leaves the approved-organizations gate in place. And regulatory risk, which is not hypothetical on this exact product: the June 12 export-control directive kept Fable 5 and Mythos 5 offline for roughly two and a half weeks before the June 30 lifting, a reminder that even a willing seller can be stopped from selling. Multiply a healthy release cadence by all three and you get something close to the 16-17 cents the year-end strike is charging. The structure mirrors the Bitcoin $100K timing ladder: the headline event feels inevitable, so the entire trade lives in the timing and the definition.

Reading The Ladder Like A Sims Output

We read a board like this the way we read a simulation distribution, not a hot take. In our DFS Sims, a player's ceiling projection matters less than the shape of the whole distribution, and the same is true here. This ladder is a survival curve: each strike is a cumulative checkpoint, and where prices climb fastest is where the market puts real launch probability. Here they barely climb at all. Month by month at the midpoints, the market hands September about 3.5%, October another 3.5%, November about 5%, and December roughly 3%. That is the flattest kind of curve, no consensus window, just slow time decay against the Yes side.

There is also an ownership dynamic DFS players will recognize instantly. Release-date markets attract an obsessive fanbase, and that flow tends to lean Yes for the reason chalk gets over-rostered in tournaments: people buy the outcome they want to watch happen, an instinct we flagged when comparing Kalshi against Underdog's pick'em boards. Volume alone cannot prove which side the crowd leaned (every contract has a buyer and a seller), but the settlement record is public: over 400,000 contracts traded on the June strike, nearly as many on July, and both paid the side that read the name-and-access test correctly. DFS players price this instinct as leverage in the NBA DataHub, where NBA DFS projections, ownership and stacks let you spot the field over-rostering a popular play relative to its actual probability. On Kalshi the price is the crowd's probability estimate, and for three straight strikes whatever hope was priced into Yes went unpaid. Our public scoreboard, the Sims vs Kalshi ledger, grades every call against how the market actually settles, precisely because "the crowd has a systematic bias here" is a testable claim, and release-date boards are one of the cleaner places to test it.

How Implied Probability And Fees Work Here

The methodology, briefly. A Kalshi contract pays $1 if the event happens, so its price in cents is the market's implied probability: 17 cents equals 17%. But Kalshi also charges a trading fee of roughly 7% of price times (1 minus price) per contract, which shifts your effective odds.

Worked Example: The 17-Cent Year-End Strike

Buy Yes at the 17-cent ask on the before-January-1 strike and the fee runs about 0.07 × 0.17 × 0.83, close to one cent per contract. Your true cost is about 18 cents, so the real probability of a public Mythos release before January 1 needs to beat roughly 18% just to break even. The fee bite is biggest in absolute cents near 50, but proportionally worst on the longshots: about 3.5% of the premium at 50 cents, 5.8% at the 17-cent strike, and 6.9% on the 2-cent September strike, the most fee-taxed row on this board. It gets worse small: Kalshi rounds each order's fee up to the next cent, so a single 2-cent contract carries a raw fee of 0.14 of a cent but gets charged a full penny, half the premium gone before the trade breathes. Stack the spread on top, 2-cent ask against a 1-cent bid, and a careless small order on the September strike is the bankroll equivalent of mass-entering a low-percentage GPP play and paying for it in units.

One honest note on what this article is not. These are live market prices plus public facts run through a projections lens, not a model output. This board has not been through our AI panel yet; the panel's price-blind verdict enrichment lands on a follow-up refresh, a treatment the 2028 Democratic nominee board already carries. Until then, treat the ladder as the market's opinion, well-informed but not gospel, and note that an eight-model panel already judged the broader AI race on a neighboring Kalshi board.

What Would Flip This Board

Now the callback, because the fine print cuts both ways. The contract terms count a model "released to only some paid members," so the Yes case does not require Anthropic to open Mythos to everyone; it requires a public paid tier that carries the Mythos name. If Anthropic ever moves Mythos 5 from approved organizations to a subscription anyone can pay for, every live strike beyond that date flips from a slow No-grind into an instant Yes settlement, and because these contracts close early the moment the event occurs, the board resolves within days instead of waiting out each date. The mirror image is the quiet killer: if Anthropic instead folds Mythos-class capability into its public Fable line and never ships the name, the capability arrives, the headlines fire, and every strike on this board still settles No. That second path is the real reason the ladder stays flat. The watch items are concrete: Anthropic launch events, changes to the Mythos access language, and any new paid tier that carries the Mythos name. Against that, the No case needs nothing to happen, the most comfortable position in any timing market but never a free one. Priced out: No on the year-end strike costs about 84 cents to win 16, roughly 19% on risk over four and a half months before fees, with the stake locked the whole way and a 16-cent Yes that only has to cash once.

That asymmetry resolves the thesis: this is a rules-reading market, not a rumor market. The crowd that lost three straight strikes was trading the announcement; the money that collected was trading the name and the access clause. That skill transfers to every timing ladder on the exchange, and it is the muscle you build projecting players instead of chasing names. For the rest of the boards we track, start at the Stokastic Predictions hub, or see where exchange pricing beats the pick'em apps in Kalshi vs PrizePicks.

Want the same distribution-first read on tonight's slates instead of next year's model launch? Try the Sims free and see the full range of outcomes, not just the headline projection.

Try the Sims Free