October AI agent market prices Anthropic at 82 percent

The October AI agent market asks which company will own the top model on a public leaderboard at a fixed afternoon check. On the Polymarket page opened for this piece, Anthropic is the leading outcome at 82%, with a Buy Yes button at 83 cents. That one-point gap is on the same screen. It is not a second market.

TL;DR - The question is "Which company has the best AI Agent end of October?" The visible outcome row says Anthropic 82%. The FAQ on that page says 83%, and the buy button says 83 cents. - CoinRithm, opened the same day, also prints Anthropic at 82%. It says that figure was up 6.0 points over 24 hours, with 24-hour volume of $1.5K and liquidity of $19.9K. - Resolution, as CoinRithm copies it from the venue, uses the Agent Arena leaderboard's Models table at 12:00 p.m. ET on October 31, 2026. AutoEval rows are ignored. This is not a trade and not a model review.

What actually happened

Polymarket lists the October AI agent event under Tech and AI. The page says the market opened August 12, 2026, at 7:52 p.m. ET. In August that clock is Eastern Daylight Time, four hours behind UTC, so the same instant is 2:52 a.m. on August 13 in Moscow. The listed end date is October 31, 2026. Total volume on the event, as printed on the page, is $56,495.

The board is a list of companies, not a single Yes. Anthropic's own line shows $27,129 in volume, 82%, Buy Yes 83 cents, and Buy No 19 cents. OpenAI's line shows $9,543, 7%, Buy Yes 9 cents, and Buy No 95 cents. Google's header chip says 2.2%, the row says 2%, and the buy button says 3.3 cents. SpaceXAI, Meta, DeepSeek, ByteDance, Moonshot, Z.ai, Alibaba, Xiaomi, MiniMax, Nvidia, Baidu, Amazon, Mistral, Meituan, and Microsoft are all on the page at about 1% or under 1%. The FAQ then says the frontrunner is Anthropic at 83% and the next outcome is OpenAI at 9%. Those FAQ numbers do not match the rows above them. A reader who screenshots only the FAQ and a reader who screenshots only the row will argue about a market that has not moved.

CoinRithm's tracker, opened October 2, lines up with the rows rather than the FAQ. It shows Anthropic 82% and calls that a 6.0 point gain over 24 hours, OpenAI 7% and up 0.5 point, Google 2% and down 0.7 point. It lists 24-hour volume of $1.5K, liquidity of $19.9K, a 2.0 point spread, and a 7-day change of minus 4.5 points. It says the data had been updated 21 minutes before the page was opened, and it scores the market 63 out of 100, which it labels medium quality. Polymarket's $56,495 is lifetime volume since August 12. CoinRithm's $1.5K is a 24-hour print. They are not rival claims about the same bucket.

The rules block on Polymarket did not expand in the page that loaded. CoinRithm prints venue terms and says they are informational. The settlement text it shows is specific. The market resolves to the company that owns the model with the highest rank on the Agent Arena leaderboard at arena.ai/leaderboard/agent, on the table under Agent Arena filtered for Models, checked on October 31, 2026, at 12:00 p.m. ET. That is 7:00 p.m. in Moscow. Models marked AutoEval at the check are out, even if they show a rank. Ties go first by which model is listed higher, then by alphabetical company names on this market. CoinRithm's FAQ says the market is scheduled to resolve on November 1, 2026. Polymarket's FAQ says on or around October 31. The page schema lists an end of 2026-11-01T03:59:00.000Z, which is 6:59 a.m. Moscow on November 1. The check time in the rules and the "resolves" label are not the same sentence. Read the check time if you care when the ranking is photographed. Read the end label if you care when the contract stops trading.

What Polymarket thinks

The crowd, on the prices above, has parked most of the October AI agent probability on one lab. Eighty-two percent on a thin book is not the same object as eighty-two percent on a market with tens of millions of dollars. This event's lifetime volume is $56,495. Anthropic's slice is $27,129. A few thousand dollars can move a number that looks like a landslide. CoinRithm's 24-hour volume of $1.5K and its 2.0 point spread say the same thing in smaller type: the quote is real, and it is not deep.

An experimental AI blurb on the Polymarket page, updated October 2, 2026, at 1:31 a.m. UTC (4:31 a.m. in Moscow), also says 82%. It talks about September releases, managed agents, and a security collaboration. The page labels that paragraph experimental, says it is not trading advice, and says it plays no role in how the market resolves. This article does not treat the blurb as the percent and does not treat it as a finding about any model. The percent used here is the outcome row, cross-checked against CoinRithm, with the 83-cent buy button and the 83% FAQ kept beside it.

What the market is actually pricing is a leaderboard snapshot, not a taste test. If the top row on that Models filter at noon ET on October 31 belongs to a different company, Anthropic's shares do not get partial credit for being close. If the top row is an AutoEval model, the rules CoinRithm prints say it is skipped. A reader who wants the story of a new model can read the Opus 5.5 note or the Grok 4.7 note. Neither of those pages is this leaderboard, and neither decides this contract.

What the two pages do not show

Neither opened page shows the current Agent Arena ranking. A price is not a rank. Neither page shows who is on the other side of the 83-cent button, or whether the 82% row and the 83-cent ask were a moment apart. CoinRithm says fees exist, the minimum order is 5 shares, and the tick size is 1 percentage point. That tick is wide enough to explain a one-point quarrel between a row and a button. It does not, by itself, explain an FAQ that says OpenAI is at 9% while the row says 7%.

The October AI agent contract also does not say the winning company has the best product for your job. The October AI agent number is a leaderboard photo, not a ranking you can paste into a budget. "Best" here is a column on one table, at one minute, with AutoEval removed and a tie-break written in advance. A lab can ship something useful and still lose this market. A lab can win this market and still be a bad fit for a private workflow.

What to do as a reader (not a trade)

This is not investment advice and not a suggestion to buy or sell any share. If you are using the number in a conversation, quote the row and the button together: Anthropic 82% on the outcome line, 83 cents to buy Yes, FAQ text at 83%. Add CoinRithm's 82% so you are not relying on one template. Say the book is small. A confident percent on $56,495 of lifetime volume is a loud room, not a deep one.

If you care who is actually first at computer-use work, open the leaderboard the rules name and read the filter. Do it before October 31, and do it again at the check, because the contract photographs one moment. Ignore the experimental paragraph when you tell someone the odds. It does not settle anything.

The pages to open are Polymarket's event and CoinRithm's tracker. The first is the venue. The second is a mirror with the rules written out and a different volume window. Where they disagree, keep both figures. Where the venue disagrees with itself, keep both of those too.