Top AI Model This Month on Prediction Markets

Anthropic breaks Google’s winning streak in the #1 AI model market

Kalshi and Polymarket both run monthly markets on which company holds the top-ranked AI model on the Arena leaderboard. Claude Opus 4.6 held the top spot on the Arena leaderboard for a second consecutive month, giving Anthropic back-to-back wins for the first time since the market launched. Both Kalshi and Polymarket resolved to Anthropic, though March was far less settled than February — Anthropic opened the month at just 54% on Polymarket with Google at 24%, the most competitive start since Google’s six-month streak began. Gemini 3.1 Pro had launched in late February just 4 Elo points behind Opus 4.6 on Arena, and GPT-5.4 dropped on March 5. Neither was enough to unseat Claude by month-end. Polymarket volume came in at $10.44M, roughly half February’s $21.66M, reflecting a market that grew more confident in Anthropic’s lead as the month progressed.

One difference worth noting: Kalshi resolves to the specific model (gemini-3-pro, claude-opus-4-6), while Polymarket resolves to the company (Google, Anthropic). Kalshi also runs both variants — a model-level market and a company-level market. Our charts are aggregating data from the model market.

Monthly AI prediction results

How to bet on AI prediction markets

Each company (or model, on Kalshi) gets its own contract. You pick Yes or No on whether that AI company will hold #1 at the end of the month. Only one can win, so in practice this is a winner-take-all contract. If you buy Anthropic Yes at 98¢ and Anthropic holds #1 on February 28, the contract pays $1.00 and every other company’s contract resolves No.

The contracts trade between 1¢ and 99¢. With only one day left in February and Anthropic at 97-99%, there isn’t much edge left to gain this month. The opportunity is in forward months where the outcome is less certain, particularly if GPT-5.3 or Grok 5 launches and disrupts the leaderboard.

The two platforms handle timing differently. Kalshi opens each month’s market on the 1st, so you can only trade the current month. Polymarket lists future months alongside the active one — March, June, and beyond are already tradeable. That means Polymarket lets you take a position on where the leaderboard will be months from now, while Kalshi is limited to the race in progress.

For a full comparison of where to trade, see our list of prediction market apps.

Anthropic beats Google for the first time in six months

Claude Opus 4.6 won March, giving Anthropic consecutive monthly titles for the first time since these markets launched. Anthropic opened at just 54% on Polymarket with Google at 24% and OpenAI at 12%, the most spread-out opening odds the field had seen in months. Gemini 3.1 Pro entered the month 4 Elo points off the pace, and GPT-5.4 dropped March 5. Neither moved the leaderboard enough to matter. Both Kalshi and Polymarket resolved to Anthropic at the March 31 snapshot.

An interesting twist, Polymarket volume fell significantly from $21.66M in Feb to just $10.44M. We saw the market start strong but teetered off by midmonth.

April could be more exciting. A March 26 data leak revealed Anthropic’s unreleased “Claude Mythos,” described as a step-change advance in coding, reasoning, and cybersecurity over Opus. Anthropic confirmed it’s in early access testing with no public release date set. Polymarket On the other side, OpenAI confirmed pre-training has wrapped on its next frontier model, codenamed “Spud,” building on GPT-5.4. Polymarket Anthropic is sitting at 95% on Polymarket for April, but either of those releases landing on Arena before April 30 changes the picture quickly.

What to watch this month

Several releases could reshape the April market:

• Anthropic’s Claude Mythos was exposed in a March 26 data leak and confirmed to be in early access with cybersecurity partners. Anthropic describes it as a step-change beyond Opus 4.6 with significant advances in coding and reasoning. No public release date is set, but if it lands on Arena in April it would be Anthropic running against itself — and likely widening its own lead.
• OpenAI’s GPT-5.5 “Spud” has completed pre-training, building on GPT-5.4’s already strong position in computer use and agentic benchmarks. OpenAI has been shipping on a near-monthly cadence in 2026. A GPT-5.5 drop that cracks the Arena top spot would end Anthropic’s streak at two.
• xAI’s Grok 4.20 full model and API is still in training. It introduces a genuinely different architecture — four specialized agents running in parallel — and full API access would open up a new class of real-time, multi-agent applications. xAI launched Beta on February 17, iterated to Beta 2 on March 3, with Beta 3 already in development, suggesting Q2 is more likely than April for a full release.

The volume trend is worth noting, too. Polymarket’s monthly AI market has been declining steadily MoM, with $36.33M in December, $28.97M in January, $21.66M in February, $10.44M in March. A competitive April with real challengers on the leaderboard could reverse that. Track live volume across Kalshi and Polymarket on our prediction markets dashboard.

How do these markets resolve?

Kalshi and Polymarket both resolve against the Arena leaderboard, but they use different columns — which means they could theoretically resolve to different winners.

• Kalshi checks the “Rank (UB)” column on the text leaderboard with Style Control removed on the last day of the month. Tiebreaker goes to highest Arena Score, then most votes, then earliest release.
• Polymarket checks the “Arena Score” column on the Leaderboard tab at 12:00 PM ET on the last day of the month with style control off. Tiebreaker is alphabetical.

Rank (UB) reflects the upper bound of a model’s confidence interval, while Arena Score is the raw Elo. In a tight race where two models are separated by a few points, one platform could resolve to a different winner than the other — use our arbitrage calculator to compare pricing across platforms.

How Arena ranks AI models

Kalshi and Polymarket both resolve against the Arena leaderboard, formerly known as the LMSYS Chatbot Arena. Arena presents users with a prompt and two anonymous model responses side by side. The user picks which response is better without knowing which model produced it. Over 5.3 million of these blind comparisons have been collected, and each model receives an Elo rating derived from its win rate against other models in the pool.

Style Control is a leaderboard filter that adjusts for formatting bias — models that use more markdown, bullet points, or longer responses tend to score higher in raw comparisons even when the underlying reasoning is equivalent. With Style Control on, Arena isolates substance from presentation. Both Kalshi and Polymarket require Style Control off in their resolution criteria, meaning the raw Elo without formatting adjustment is what determines the winner.

Related AI prediction markets

• “Which companies will have a top-ranked AI model this year?” on Kalshi. An annual market with 12 active contracts ($832K volume). Unlike the monthly markets, these resolve Yes if a company reaches #1 at any point during 2026. Google and Anthropic have already resolved Yes. OpenAI leads the remaining contracts at 62%, xAI at 55%.
• “AI capability growth before July?” on Kalshi. A related bet on whether any model will hit an Arena Score of 1550 or higher before July ($77K volume). Currently pricing 46% for 1550 and 30% for 1575. Claude Opus 4.6 sits at 1506 today, so this is effectively a bet on the next generation of models.
• “When will Claude 5 be released?” on Polymarket. Currently pricing 36% for April 30. If Claude 5 arrives and tops Arena, Anthropic could hold #1 for consecutive months.
• “When will GPT-5.3 be released?” on Polymarket. Directly relevant to whether OpenAI can challenge for March. If GPT-5.3 launches and climbs Arena, expect the monthly market to move in tandem.
• EU AI Act compliance deadline arrives August 2, 2026. If a lab delays a frontier model release to meet compliance requirements, that’s one fewer chance to top the leaderboard within a given month.

Liquidity varies significantly across these markets. Polymarket’s monthly AI market routinely draws $20-36 million per month, making it one of the deeper non-political betting markets on the platform. Kalshi’s monthly markets are thinner at $2-3 million but growing — February volume nearly doubled January. The annual and milestone markets on either site are much smaller, typically under $1 million, so larger positions may face slippage.

The fee structures are quite different between Kalshi and Polymarket and can affect net returns, particularly on lower-volume contracts. If you’re new to betting on Kalshi or Polymarket, you will also want to learn how order books work and market resolution before making your first trade.

FAQ

What does “best AI model” mean in these markets?

The company (or model) that holds the top rank on the Arena text leaderboard at the end of the month. Kalshi uses the Rank (UB) column, Polymarket uses Arena Score.

Can Kalshi and Polymarket resolve to different winners?

In theory, yes. They use different columns from the same leaderboard. In practice, the #1 model under Rank (UB) and Arena Score has been the same, but a tight race could produce a split.

Why is Polymarket’s volume so much higher than Kalshi’s?

Polymarket generally carries more liquidity on AI-related markets and draws a larger crypto-native trading audience. Kalshi is a regulated US exchange that has been expanding internationally, while Polymarket US continues to roll out to domestic traders.

Why is Anthropic favored for April but not dominant?

Anthropic opened April at 54% because Claude Opus 4.6 is the incumbent — it holds Arena #1 at 1506 Elo with a large gap over Gemini 3 Pro. Google sits at 24% partly because Gemini 3.1 Pro launched on February 19 and is already within 4 Elo points of Opus 4.6 on Arena. As more votes accumulate through March, that gap could close. OpenAI is at 12% because the general GPT-5.3 model is expected in Q1 2026 — the coding variant (GPT-5.3-Codex) shipped in early February, but the full model hasn’t launched yet. If it arrives in March and climbs Arena, OpenAI could become a real contender. xAI rounds out at 5.5% with Grok 5 confirmed for Q1 2026, though a public beta likely lands closer to March or April.

How often do these markets run?

Monthly on Kalshi and Polymarket. New contracts open as the prior month’s market approaches resolution. Polymarket has results going back to at least August 2025, Kalshi to at least August 2025 as well.