Anthropic is winning the AI model race on Polymarket — and OpenAI knows it
On 10 June 2026, a contract on Polymarket put the chances of Anthropic holding the best frontier AI model at the end of the month at 91%. A second contract, posted the previous evening, gave Anthropic a 72% probability of going public before OpenAI. A third market, opened 9 June, gives GPT-5.6 a 68% chance of arriving next week — described in the market's own framing as OpenAI's response to Anthropic's Claude Fable launch. Three numbers, one story: the prediction market has, in the space of 48 hours, repriced the AI industry's hierarchy.
The signal is not subtle. When retail and professional bettors, moving real money, put a 91% probability on a single outcome, the phrase "strong favourite" stops being metaphor. The contract, hosted on the prediction-market platform Polymarket, is the cleanest read yet on a contest the frontier-model coverage has tended to treat as neck-and-neck.
What the contracts are actually saying
The three markets are not the same instrument and should not be read as one. The end-of-June "best AI model" contract, at 91%, is a snapshot of the present — a bet that Anthropic's current top-tier system will still be the most capable on 30 June. The IPO contract, at 72% in Anthropic's favour, is a bet on corporate-finance choreography: who files, who prices, who lists first. The GPT-5.6 release-date market, at 68% for "next week", is a bet on a specific defensive move by OpenAI to blunt the Claude Fable launch.
Taken together, the picture is of an incumbent position under pressure. Anthropic is not just winning the model race this month; it is on track to monetise the lead in the public markets before its better-capitalised rival. The IPO contract is the more consequential of the two — capital-structure beats benchmark — and it has moved sharply in Anthropic's direction.
The product underneath the bet
The product generating the noise is Claude Fable 5. On 9 June 2026, TechCrunch reported that Anthropic's latest model "can make weirdly fun video games with the click of a button" and that it is "going to be a big hit with the web's vibe coders." That description matters more than it sounds. The frontier-model contest of 2024 and 2025 was largely a text-and-reasoning contest, judged on benchmarks and chatbot feel. Fable 5, by the framing TechCrunch is using, is a generative-product contest: ship a small, complete, playable artefact in one prompt. That is a different surface area, and it is one the incumbent chatbot narrative does not cover well. The "vibe coder" framing — non-engineers shipping interactive products — is the cultural seam OpenAI has been trying to sew into ChatGPT since the start of the year. Anthropic, on the strength of this coverage, has beaten OpenAI to the seam.
The counter-read
The counter-narrative is straightforward: prediction markets price narratives, not ground truth. A contract at 91% is, by construction, less informative than it looks — 9% of the float is still willing to bet against the favourite, and the historical record of Polymarket contracts settling on the right side of headline consensus is mixed. A second objection: the IPO contract is sensitive to legal and regulatory friction, not just corporate intent. OpenAI's structure, with its capped-profit parent and non-profit overseer, has historically complicated a clean IPO path, but that is precisely the kind of friction that can resolve quickly, in either direction, on a single board decision. The market may be pricing the friction rather than the lead.
A third, more structural read: GPT-5.6, if it ships next week, changes the underlying condition. A 68% probability is a hedge, not a fait accompli, and a successful OpenAI release could compress both Anthropic contracts inside 72 hours. The market is, in effect, telling the reader that the next ten trading days are the window.
The structural frame
The interesting shift is not the model race itself — frontier leads change hands every quarter — but the speed at which the prediction-market layer is absorbing the news. Polymarket's contracts on AI leadership have been, until this month, comparatively thin and noisy. The triple-stack of high-conviction contracts on Anthropic, Fable 5, and GPT-5.6 suggests that liquidity is arriving. When prediction markets become the cleanest read on a technology race, the technology race has crossed a line from R&D story to financial-asset story. The capital-markets desk is now a co-author of every product launch.
Stakes
If Anthropic closes June still on top, and if the IPO contract resolves in its favour before OpenAI lists, the resulting asymmetry is large. Anthropic becomes the public-market vehicle for the AI trade, with the disclosure regime, the analyst coverage, and the employee-liquidity mechanics that go with it. OpenAI, in that scenario, is the private incumbent raising inside rounds at a discount to a public peer. The boardrooms in San Francisco are repricing their option grants accordingly; the wire desks are simply late to the print.
What remains uncertain is the durability. The end-of-June contract is a one-month read. The IPO contract is a 2026-to-2027 read. The GPT-5.6 release-date market is a one-week read. They reinforce each other today, but a single strong OpenAI release in the next ten days would knock all three off their pegs. The market is pricing Anthropic as the favourite. The market is not, yet, pricing the possibility that the favourite changes.
Desk note: Monexus read the three Polymarket contracts as a single signal — the prediction-market layer has absorbed the Fable 5 news faster than the wire desks have. We hold the line that the underlying product still has to ship, and the IPO contract is sensitive to regulatory path, not just technical lead.