There is a peculiar honesty to a contract that hands the resolution job to a public leaderboard. No expert panel. No vibe check. Just a screenshot of a website on a specific afternoon, and whoever sits at the top of one particular column takes the pot.
That is the shape of Polymarket's Which company has best AI model end of June? market, which settles on 30 June 2026 at 12:00 PM ET against the Chatbot Arena leaderboard at lmarena.ai. Fifteen named labs are on the ticket, from Anthropic and Google through to Moonshot and Baidu. With the deadline now close, the interesting question is not who is in front today; it is what this contract is really measuring, and whether the thing it measures is the thing you think it is.
What this market actually resolves on
The resolution rule is unusually crisp for an AI market. At noon Eastern on 30 June, whoever checks the contract opens lmarena.ai, clicks the Leaderboard tab, selects the text leaderboard, and turns style control off. The model sitting at Rank 1 wins the contract for the company that owns it. Ties get broken by the Arena score underneath the rank.
That is the whole mechanism. No averaging across benchmarks. No weighting by parameter count, training cost, or release recency. No carve-out for whether the top model is a frontier flagship or a tuned variant of something a quarter the size. The contract is a snapshot of one ranking, on one site, at one moment.
Which means the things this contract is good at telling you and the things it is bad at telling you diverge quite sharply. It is a clean read on Chatbot Arena head-to-head preference at noon ET on a Tuesday. It is not a clean read on which company has the smartest model, the most useful model, the most-shipped model, or the model your engineering team would actually choose. Arena ranks correlate with capability. They are not capability.
This matters because Arena's methodology has its own texture. Pairwise blind voting from a user pool rewards models that produce confident, well-formatted, instruction-following answers to the kinds of prompts users tend to type. A model that is genuinely stronger on long-horizon reasoning or coding can still sit behind a model that is friendlier on short turns. That is a feature of Arena, not a bug, but it is worth holding in mind before reading the contract as a verdict on AI supremacy.
How to read the snapshot
The price board as of 15 June leans hard in one direction. Anthropic sits around 87%, with Google near 8% and OpenAI near 6%, as of 15 June 2026. Everyone else, from xAI to DeepSeek to the Chinese cluster of Z.ai, Moonshot, Meituan, Alibaba, ByteDance and Baidu, trades below one percent.
That is a clear favourite, not a contested race. The market is pricing Anthropic as the overwhelmingly likely owner of whichever model is sitting at Rank 1 when the screenshot gets taken. Worth flagging though: the contract's structure means the level can look more decisive than the underlying uncertainty really is. With two weeks to go, a single new release from Google, OpenAI or any of the trailing labs that lands at the top of Arena before noon ET on the 30th flips the entire contract. The contract does not care whether the lead has been held for a day or a month.
That is the qualitative point: a leaderboard contract has high sensitivity to late-arriving entrants by design. A long-running political market can absorb a surprise; a single-day leaderboard snapshot cannot. The favourite has the favourite's price for a reason, but the price is paying for the absence of a late top-of-table release as much as for current standing. If you want to think about that structure properly, our explainer on how prediction market odds work and the companion piece on what implied probability actually tells you are the better starting points than the price ticker itself.
Why the contract exists in this shape
There is a reason designers reach for an external leaderboard rather than a panel or a benchmark composite. Resolution sources need to be public, timestamped, and uncontroversial; otherwise the resolution itself becomes the contested event. Chatbot Arena is publicly hosted, continuously updated, and the top-of-table entry is a single row anyone can read. That is roughly the gold standard for an event contract resolution source: minimal interpretation, minimal lag, minimal ambiguity.
The trade-off is what we have already covered. The cleanliness of the source is bought at the cost of measuring exactly what the source measures, and nothing more. A contract resolved by Arena rank is a contract about Arena rank. Read it as such, and it is a useful and well-built instrument. Read it as a referendum on which lab has the best AI, and you are reading more into the price than the contract is built to carry.
Which is the broader point about specialist event contracts more generally. The good ones do one thing precisely. The trader's job, and the reader's, is to know which thing.
iPredicta tracks the AI-leaderboard contracts across Polymarket and the regulated US venues, and the 30 June resolution on this one is exactly the sort of single-source, single-moment settlement worth watching closely, both for what it pays out and for what it reveals about how these contracts behave as the deadline closes in.
Frequently asked questions
What happens if a new model launches and tops Chatbot Arena on 30 June itself?
The contract reads whoever is at Rank 1 at noon ET on 30 June, with style control off. If a new release lands at the top of the leaderboard hours before the check, that release's owner takes the contract. The mechanism does not care how long the lead has been held, which is precisely why a single-day snapshot contract can be sensitive to late entrants.
Does this contract tell me which company has the best AI model overall?
Strictly, no. It tells you which company owns the model sitting at Rank 1 on Chatbot Arena's text leaderboard at one specific moment. Arena ranks reflect blind pairwise user preference, which correlates with capability but is not the same thing as it. Treat the contract as a precise read on a specific leaderboard, not a verdict on AI supremacy.
