Best AI model odds: Google is 70% to top the leaderboard at the end of 2026, and Anthropic has fallen from 72% to 22%

The prediction markets have turned decisively towards Google in the race for the best AI model of 2026. Polymarket’s market on which company will have the top model at the end of the year has Google at 70.5% and Anthropic at 22%, with OpenAI at 2.8%. On Kalshi, which lists the same question by product name, Gemini is at 64.9¢ and Claude at 29.2¢, with ChatGPT at 3.7¢.

Two months ago it was the other way round. In mid-September Polymarket had Anthropic at 72.5% and Google at 9.5%. The two crossed in October, and the gap widened sharply on 10 November: Google rose from 58.5% to 70.5% and Anthropic fell from 36% to 22% within a few hours.

The odds

Polymarket: midpoint or last price; Kalshi: last trade. 10 November 2026.

Kalshi’s market is the busier of the two by contract count. Claude’s contract has traded about 3.6 million contracts, Gemini’s 3.3 million and ChatGPT’s 3.3 million. Polymarket’s end-of-2026 market has traded $4.18 million since it opened on 4 June.

The two exchanges disagree by about six points on Google, 70.5% against 64.9%. Part of that is timing: Polymarket moved further on 10 November. But it is also a reminder that thin markets can sit apart for a while (Kalshi v Polymarket price gaps).

How the odds flipped

For most of the summer the market was a one-company race. Anthropic was priced between about 66% and 72% from June to mid-September, while Google was 8.5% to 14% and OpenAI about 10%.

The turn came at the start of October. Between 15 September and 1 October, Google jumped from 9.5% to 38.5% and Anthropic fell from 72.5% to 50.5%. Google passed Anthropic in October and kept climbing: 56.5% on 1 November, 65% on 9 November and 70.5% on the afternoon of 10 November.

The monthly markets tell the same story. Anthropic topped the leaderboard at the end of every month from February to September 2026, eight in a row, according to Polymarket’s settled monthly markets. Google won October. For November, Polymarket has Google at 75.5% and Kalshi has Gemini at 74¢ and Claude at 23¢ (our study of the monthly markets).

The monthly scoreboard

Across the 20 monthly markets that have settled since February 2025, Google has won 11, Anthropic 8 and xAI 1. OpenAI has not won any, although it was the narrow favourite a month before the end of August 2025.

The pattern is one of long reigns. Google won every market we found from March 2025 to January 2026. Anthropic took over in February 2026 from a price of about 3% a month out and held the top spot through September. Google won October 2026 at 98% a day before the check. On that record, the end-of-year market is asking whether Google’s new reign lasts for at least two more months.

Kalshi’s version: products, not companies

Kalshi’s market lists products rather than companies: Gemini, Claude, ChatGPT, Grok, Qwen, Ernie, Kimi and Muse. It resolves on the company with the top-ranked model, so in practice Gemini stands for Google, Claude for Anthropic and ChatGPT for OpenAI. The naming changes nothing about the outcome, as long as you map each product to its company correctly, but it does mean the same question trades on two platforms with different labels.

The Kalshi market is notable for its trading in long shots. The contracts on Muse and Grok, priced at about 1¢, have each traded more than 2 million contracts, and Kimi, Qwen and Ernie more than 800,000 each. That is a lot of activity on outcomes the market prices at 1% or less.

Kalshi also runs monthly and weekly versions. Its November market has Gemini at 74¢ and Claude at 23¢, very close to Polymarket’s 75.5% for Google in its November market.

How the markets settle

Both markets settle on the Text Arena leaderboard, the crowd-sourced ranking of chatbots formerly known as LMArena (Chatbot Arena), now at arena.ai. Users compare two anonymous models side by side and vote for the better answer, and the votes produce a score for each model.

• Polymarket: the company whose model has the highest rank on the text leaderboard with style control off, checked on 31 December 2026 at 12:00 p.m. ET. Ties on rank are broken by Arena score, including unrounded values, then alphabetically. The monthly markets use the same rules, check the “Text Arena (Overall)” table and exclude models marked “AutoEval”.
• Kalshi: the company with the top-ranked model on 31 December. Ties on rank go to the higher Arena score, then the model with more votes, then the one released earlier. Kalshi tells traders to switch on the leaderboard’s “Remove Style Control” toggle when checking.

The style-control setting matters. Arena publishes rankings with and without an adjustment for answer length and formatting, and the two can have different leaders. Both markets use the version without style control. Before trading, check the leaderboard in exactly the view the rules name (how to read market rules).

What the monthly markets teach

Polymarket has run a monthly version of this question since early 2025, and 20 of them have now settled. The record is useful for anyone trading the year-end market:

• A day before the check, the favourite won every time, 20 out of 20, at an average price of about 98%. By then the leaderboard rarely changes.
• A month out, the favourite won 16 of 20 at an average price of 74%. Four times the favourite a month before the end lost: in February 2025, when xAI won from a price of about 2%; in March 2025 and August 2025, when Google came from behind; and in February 2026, when Anthropic won from about 3% a month out.
• Leads tend to last. Google won every monthly market we found from March 2025 to January 2026, and Anthropic won eight in a row from February to September 2026.

With seven weeks to go, the year-end market is roughly where the monthly markets have been a month or so before the check. At that distance, the favourite has won about four times in five, and long shots have occasionally come through.

What would move the market

• New model releases. A new model that reaches the top of the text leaderboard would change the result. Any release from Google, Anthropic or OpenAI before the end of December is the biggest risk to the current prices.
• Leaderboard updates. Scores change as votes come in, so the order near the top can shift even without a new release.
• The check moment. The market settles on the table at noon ET on 31 December, not on an average over the month.

What the price implies

A 70.5% price for Google seven weeks before the check is lower than the favourite’s typical price a week or a day out in the monthly markets, but higher than its typical price a month out. In the 20 monthly markets, the favourite a month before the check was priced at about 74% on average and won 80% of the time; two weeks out it averaged 88% and won 90%.

Read that way, the year-end market looks about where you would expect for this distance from the deadline. The one thing that would move it decisively is a new model at the top of the board. If Google still leads in mid-December, history suggests the price would be in the 90s; if Anthropic or another company takes the lead, the monthly markets show how fast prices follow: Anthropic went from about 3% to 73% in two weeks in February 2026.

The separate prices on Polymarket’s November market, 75.5% for Google, and the end-of-year market, 70.5%, also tell you something. The extra month of uncertainty to the end of December is worth about five points to traders.

How to trade it

Buying Google at 70.5¢ pays $1 if a Google model is top of the text leaderboard (style control off) at noon ET on 31 December, a return of about 42% if right. Buying Anthropic at 22¢ pays about 4.5 times the stake. Both platforms charge only takers: Polymarket’s fee on this market is 4% × price × (1 − price) per share, and Kalshi’s standard fee is 7% × price × (1 − price) (fees compared).

The main risk is a release you cannot predict, and the market gives you no warning: a new model can appear on the leaderboard between one check of the price and the next. Size positions so that a surprise launch does not hurt you more than you can afford (Kelly criterion).

FAQ

Who has the best AI model right now? The markets imply Google: Polymarket gives it about 75% to be top at the end of November and 70% at the end of the year, and Google’s model topped the leaderboard at the end of October according to Polymarket’s settled October market.

What leaderboard do the markets use? The Text Arena leaderboard (formerly LMArena), overall ranking, with style control off.

When do the markets settle? Polymarket checks at 12:00 p.m. ET on 31 December 2026; Kalshi settles on the top-ranked model on 31 December.

Does it matter which model, or only which company? Only the company. Any Google model at the top settles the market for Google, whichever version it is.

Has OpenAI ever won one of these markets? Not in the 20 monthly markets that have settled on Polymarket. It was the narrow favourite a month before the end of August 2025, at 51%, but Google won.

Where can I see the leaderboard? On the Text Arena leaderboard at arena.ai. Make sure style control is switched off, as both markets require.

Live prices: Polymarket’s end-of-2026 market, its November market and Kalshi’s.

Prices as of 10 November 2026. Sources: Polymarket and Kalshi market data and rules. Nothing here is financial advice.