# Best AI Model Odds: Polymarket's Favourite a Week Out Won 21 of 22 Month-End Checks
Polymarket asks every month which company will top the LMArena text leaderboard. A week before each check from December 2024 to September 2026, the favourite won 21 times in 22. The one miss: Gemini 2.5 Pro, released six days before the March 2025 check.
Canonical URL: https://vultax.com/research/best-ai-model-odds-polymarket
Author: [Vultax Research](https://vultax.com/editorial-policy)
Published: 2026-10-26
Last public revision: 2026-10-26
AI-assisted drafting, research and editing. Published by OmniOS OÜ. [How we use AI](https://vultax.com/editorial-policy#ai)
Published research snapshots and calculations; these exports do not contain current quotes or live account data. Observation windows and populations are stated in the source captions and methodology. Missing metadata is unknown, not zero. A reading of public market prices at stated times, scored against settled results; not a forecast by Vultax, not a ranking of AI models and not advice to trade.
## Key figures
- Month-end checks the week-out favourite won: 21 of 22 — Polymarket's monthly best AI model markets, December 2024 to September 2026. The miss was March 2025, when Gemini 2.5 Pro arrived six days before the check
- xAI a week before the March 2025 check: 92.5% — Google was 5.2%. Google released Gemini 2.5 Pro on 25 March and won the 31 March check
- Month-ends with Anthropic on top: 8 — Every check from February to September 2026. Google held the top spot at 13 month-ends before that, xAI at 1
- Google for October 2026, read 19 October: 73.1% — Anthropic 25.5%, at 21:16 UTC on 19 October 2026, twelve days before the check
## The short answer
Polymarket runs a market every month on which company will have the top model on the LMArena text leaderboard when it is checked at 12:00 Eastern on the last day. Over 22 month-end checks from December 2024 to September 2026, the company with the highest price a week before the check won 21 times. A day before, the favourite won all 22.
The one miss came from a release. On 24 March 2025, a week before the check, xAI was 92.5% to hold the top spot at the end of the month and Google was 5.2%. Google released Gemini 2.5 Pro on 25 March, it went straight to first place, and between 14:22 and 16:57 UTC that day Google's price went from 15.0% to 86.4%.
Three companies have held the top spot at a month-end: Google 13 times, Anthropic 8 times in a row from February 2026, and xAI once, in February 2025. Prices are Polymarket's public Yes prices; the winners are the markets' settled results.
## Every month-end check since December 2024
The table shows the favourite and its price 14 days and 7 days before each check, and the winner's price a day before. The median price of the eventual winner a week out was 96.3%.
The markets have changed their wording over time, from the model to the company, and their source, from lmarena.ai to arena.ai. Each settled on the top model of the text leaderboard with style control off at its stated check time.
Polymarket Yes prices at the minute 14 days, 7 days and 1 day before each 12:00 Eastern check. 'Not open': the market opened later. The December 2024 and January 2025 markets named models (Gemini, Grok, ChatGPT, Claude); they are shown by company.
| Month-end | Top company | Favourite 14 days before | Favourite 7 days before | Winner, 1 day before |
| --- | --- | --- | --- | --- |
| December 2024 | Google | not open | Google 86.5% | 99.4% |
| January 2025 | Google | Google 90.0% | Google 94.3% | 98.8% |
| February 2025 | xAI | Google 81.0%, missed | xAI 79.9% | 86.1% |
| March 2025 | Google | xAI 86.0%, missed | xAI 92.5%, missed | 98.2% |
| April 2025 | Google | Google 63.5% | Google 77.0% | 98.2% |
| May 2025 | Google | Google 87.5% | Google 91.0% | 97.9% |
| June 2025 | Google | Google 92.0% | Google 94.5% | 99.4% |
| July 2025 | Google | Google 91.0% | Google 94.0% | 97.7% |
| August 2025 | Google | Google 93.5% | Google 97.0% | 99.6% |
| September 2025 | Google | Google 95.7% | Google 98.7% | 99.5% |
| October 2025 | Google | Google 96.8% | Google 97.6% | 99.6% |
| November 2025 | Google | Google 94.3% | Google 98.2% | 99.8% |
| December 2025 | Google | Google 95.5% | Google 97.1% | 99.2% |
| January 2026 | Google | Google 94.8% | Google 98.4% | 99.8% |
| February 2026 | Anthropic | Anthropic 68.3% | Anthropic 90.0% | 98.8% |
| March 2026 | Anthropic | Anthropic 92.8% | Anthropic 95.6% | 99.8% |
| April 2026 | Anthropic | Anthropic 93.5% | Anthropic 69.0% | 99.2% |
| May 2026 | Anthropic | Anthropic 91.5% | Anthropic 98.1% | 99.7% |
| June 2026 | Anthropic | Anthropic 91.6% | Anthropic 98.2% | 99.6% |
| July 2026 | Anthropic | Anthropic 96.8% | Anthropic 98.8% | 99.9% |
| August 2026 | Anthropic | Anthropic 97.3% | Anthropic 98.3% | 99.7% |
| September 2026 | Anthropic | Anthropic 96.2% | Anthropic 98.5% | 99.5% |
[CSV](https://vultax.com/research/best-ai-model-odds-polymarket/data.csv?dataset=every-month-table), [JSON](https://vultax.com/research/best-ai-model-odds-polymarket/data.json?dataset=every-month-table)
## March 2025: the week Gemini 2.5 arrived
For most of March 2025 the market expected xAI to keep the top spot it had taken in February. In the midnight UTC readings xAI rose from 54.5% on 11 March to 92.5% on 23 March, while Google fell from 6.5% to under 1%. Google climbed to 13.1% late on 24 March. On 25 March it published Gemini 2.5 Pro and said the model debuted at first place on LMArena by a significant margin. Between 14:22 and 16:57 UTC that afternoon Google's price went from 15.0% to 86.4% and xAI's fell from 81.0% to 12.0%.
It is the only month in the record where the favourite a week before the check lost. The market had no way to price a release it did not know about; what it did was move within a day once the leaderboard changed.
### Best AI model at the end of March 2025: xAI and Google
Polymarket, 00:00 UTC readings, 1 to 31 March 2025. Percent.
| Category | Google (%) | xAI (%) |
| --- | --- | --- |
| 1 Mar | 10.5 | 46.5 |
| 2 Mar | 8 | 56.5 |
| 3 Mar | 9 | 60 |
| 4 Mar | 8.5 | 53.5 |
| 5 Mar | 7 | 50.5 |
| 6 Mar | 5.5 | 51.5 |
| 7 Mar | 5 | 51.5 |
| 8 Mar | 6.5 | 56 |
| 9 Mar | 6 | 58 |
| 10 Mar | 7 | 54.5 |
| 11 Mar | 6.5 | 54.5 |
| 12 Mar | 6.5 | 54.5 |
| 13 Mar | 6.5 | 67 |
| 14 Mar | 6.5 | 73.5 |
| 15 Mar | 6 | 75.5 |
| 16 Mar | 6.5 | 77.5 |
| 17 Mar | 5.5 | 84.5 |
| 18 Mar | 4.5 | 86.5 |
| 19 Mar | 5 | 89 |
| 20 Mar | 2 | 88.5 |
| 21 Mar | 0.7 | 90.5 |
| 22 Mar | 0.7 | 91 |
| 23 Mar | 1 | 92.5 |
| 24 Mar | 4 | 91 |
| 25 Mar | 13.1 | 80.5 |
| 26 Mar | 94.3 | 4.2 |
| 27 Mar | 95 | 1.4 |
| 28 Mar | 97 | 2.2 |
| 29 Mar | 97.4 | 1.7 |
| 30 Mar | 98 | 0.8 |
| 31 Mar | 98.7 | 0.4 |
[CSV](https://vultax.com/research/best-ai-model-odds-polymarket/data.csv?dataset=march-2025-chart), [JSON](https://vultax.com/research/best-ai-model-odds-polymarket/data.json?dataset=march-2025-chart)
## The other changes at the top
February 2025 was the first turn. Two weeks before the check Google was 81.0% and xAI 6.4%. xAI unveiled Grok 3 that week; its announcement, dated 19 February, said an early version had topped the leaderboard. In the midnight UTC readings xAI's price went from 19.9% on 18 February to 77.8% on 19 February. A week before the check it was the favourite at 79.9%, and it won.
Anthropic took the top spot at the end of February 2026. Two weeks before that check it was already the favourite at 68.3%, and 90.0% a week before. It has won every check since. The closest call was April 2026: at midnight UTC on 23 April OpenAI was 28.1%, and Anthropic was 69.0% a week before the check, before it went back above 90%.
## October and the end of 2026
The October 2026 market is checked at 12:00 Eastern on 31 October. At 21:16 UTC on 19 October, twelve days before, Google was 73.1% and Anthropic 25.5%, so the market expected a change at the top for the first time since February. The market for the end of 2026 had Anthropic at 51.5%, Google 40.5% and OpenAI 5.9%.
Those readings were taken before this page was published and are kept as they were, so they can be checked against the results. The live prices are in the AI race room.
## How to read best AI model odds
Four things to keep in mind when reading the monthly market.
- The market is about one leaderboard. It settles on the top model of the LMArena text ranking with style control off, which is a measure of human preference in head-to-head chats, not of every kind of ability.
- New releases move it. Both changes the market missed or caught late came from a model released within two weeks of the check.
- A week out is informative, two weeks less so. The favourite 14 days before the check lost twice in the record, in February and March 2025.
- The company is what counts. A new model from the company already on top does not change the outcome.
## Method and limits
The record is Polymarket's 22 monthly markets on the top AI model or company, from the check on 31 December 2024 to the check on 30 September 2026, each read at the minute 14 days, 7 days, 1 day and 1 hour before 12:00 Eastern on its check day. The winner is the market that settled Yes. Brands in the two earliest markets, Gemini, Grok, ChatGPT and Claude, are counted as Google, xAI, OpenAI and Anthropic.
The August 2026 market settled on 28 September, four weeks after its check, so its settlement time is not used here. Release dates come from the companies' own announcements.
22 months are a short record and most of them were held by one company. This page reads public prices; it is not a forecast by Vultax, not a ranking of AI models and not advice to trade.
## Questions
### Which company has the best AI model right now?
At the September 2026 check, Anthropic, for the eighth month in a row on the LMArena text leaderboard. For October, Polymarket had Google at 73.1% and Anthropic at 25.5% at 21:16 UTC on 19 October 2026.
### How accurate are Polymarket's best AI model markets?
A week before each month-end check from December 2024 to September 2026, the favourite won 21 of 22 times. A day before, it won all 22.
### How is the best AI model market settled?
By the company that owns the top model of the LMArena text leaderboard, overall, with style control off, when it is checked at 12:00 Eastern on the last day of the month.
### Which company will have the best AI model at the end of 2026?
At 21:16 UTC on 19 October 2026 Polymarket had Anthropic at 51.5% and Google at 40.5%. The market is checked on 31 December 2026.
## Revision notes
- 2026-10-26: Published five days before the October 2026 check, with the October and year-end markets read at 21:16 UTC on 19 October.
## Sources and methodology
- [Polymarket — Which company has the best AI model end of October?](https://polymarket.com/event/which-company-has-the-best-ai-model-end-of-october) — Read at 21:16 UTC on 19 October 2026.
- [Polymarket — Which company has best AI model end of 2026?](https://polymarket.com/event/which-company-has-best-ai-model-end-of-2026) — Read at 21:16 UTC on 19 October 2026.
- [Polymarket — Which company has best AI model end of March? (2025)](https://polymarket.com/event/which-company-has-best-ai-model-end-of-march) — Settled 31 March 2025. The other monthly markets follow the same pattern.
- [Polymarket API — price history](https://docs.polymarket.com/) — Minute and daily Yes prices from the CLOB API. Vultax computed the checkpoint readings.
- [Arena — Text leaderboard, style control off](https://arena.ai/leaderboard/text/overall-no-style-control) — The resolution source named in the 2026 markets.
- [Google — Gemini 2.5: Our most intelligent AI model](https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/) — 25 March 2025.
- [xAI — Grok 3 Beta](https://x.ai/news/grok-3) — 19 February 2025.
- [Conditional formula proofs](https://vultax.com/datasets/research-proof-library-2026-10-25/scopes/692a1b5137d365feb78981f6187a79a9a32da42549b1bd4a6f82fa688413eea6.md) — Conditional formulas only. The proofs cover the conditional mathematical formulas; article-specific measurements and numerical results have not been proved.
- [Per-article scope records and Lean package](https://vultax.com/datasets/research-proof-library-2026-10-25/README.md) — Dated per-article scope records, Lean source, frozen permitted public inputs, checker logs and a reproducible ZIP. Covers only the scope stated in each record, not the whole study.