"Best AI Model 2026: Markets Say Google Wins October, Anthropic Wins the Year"
Ask prediction traders who has the best AI model and you will get two different answers, depending on which date you ask about. Check the leaderboard on October 31 and the answer is Google, almost certainly. Check it again on December 31 and the answer flips to Anthropic. That contradiction is the most interesting thing I found in the markets this week, because both answers come from the same traders, pricing the same leaderboard, only two months apart.
The question is narrower than it sounds, and that narrowness is the whole story. Polymarket's "best AI model" markets do not ask which lab is best in general. They resolve on one specific table: the LMArena Text Arena (Overall) leaderboard, with style control off, as it stands at 12:00 PM ET on the check date, excluding models marked "AutoEval". So the crown is not a philosophy or a product review. It is a rank on a table on a day, decided by blind human votes.
On the October table, Google's lead looks close to settled. As of October 28, Polymarket priced Google at 95.7% to hold the top spot on October 31, against 3.2% for Anthropic and a fraction of a percent for OpenAI. Kalshi's equivalent market for October 31, as reported on October 28, told the same story: Gemini 4 Argon at about 92%, Claude Opus 5.5 at about 6%. The leaderboard behind those prices shows why. Google's Gemini 4 Argon (high) sits first at 1517, ahead of Anthropic's Claude Opus 5.5 (high) at 1509, with Meta's Muse Wonder also at 1509. Argon arrived at the end of September and flipped the October market within hours; over the past week alone, Google's October price rose from about 81% to 96%, according to Polymarket daily closes. An eight-point lead with three days to go is a very steep hill for anyone else.
| Market check date | Anthropic | OpenAI | |
|---|---|---|---|
| October 31, 2026 | 95.7% | 3.2% | 0.3% |
| November 30, 2026 | 50.0% | 45.0% | 3.4% |
| December 31, 2026 | 34.5% | 56.0% | 5.9% |
The November market is where the story turns. There, Google leads Anthropic by only five points, 50% to 45%, a coin flip with a slight lean. And by December 31, the market has Anthropic ahead outright, 56% to 34.5%, with OpenAI a distant third at 5.9%. Kalshi's end-of-2026 board agrees in direction: about 59% for Claude against about 30% for Gemini, as shown on October 28. Two venues, same shape: Google now, Anthropic later, November as the hinge.
Why would traders expect the lead to change hands? Part of the answer is how thin Argon's lead is. In a mid-October snapshot, Argon was still marked preliminary, with only about 4,900 votes behind its score, against about 8,300 for Opus 5.5. Preliminary scores move as votes accumulate, and they do not always move up. Another part is breadth. Anthropic does not just have one horse in this race: it holds the top of several other boards at once, including Artificial Analysis' intelligence index, LiveBench, LMArena's WebDev board, and the agent and vision boards, while OpenAI leads on the hardest academic questions. None of that decides the December check, which uses only the text board, but it means Anthropic has more than one model within striking distance if any single score shifts. And there are still two months of releases to come; Anthropic also filed to go public this week, which puts its next move under more scrutiny, not less.
I am not as confident as the December market, and the reason is the board itself. The December check is decided on the one table Google currently leads, and no new Anthropic model has been announced for the run-up to year-end. If Argon's score holds as votes pile in, and if Anthropic's next release slips into 2027, Google keeps the crown and the market's flip never happens. That is why I put my own number below the market's: I lean Anthropic, but only just.
My prediction, then, is that Anthropic will own the top-ranked model on the LMArena Text Arena (Overall) leaderboard, style control off, when it is checked on December 31, 2026. I put that at 50%, against the market's 56% for Anthropic on October 28. The November check is the early verdict to watch: if Anthropic has not taken the lead by then, the case for December gets much weaker. And whatever happens, remember what is actually being measured. A leaderboard crown is a snapshot of human preference on one table on one day, not a verdict on which lab builds the best AI. That is why the market can crown two different winners in two months.