Which company has the best AI model end of September?

Platform comparison

Market context

The Arena.ai Text Arena leaderboard will determine which company operates the highest-ranked large language model as of 30 September 2026. The resolution mechanism relies on a single snapshot of the "Overall" rankings table with style control disabled, meaning models are evaluated on raw performance metrics without stylistic adjustments. This creates a binary outcome: one company's model will occupy the top position, and that entity will be deemed the winner regardless of margin.

Historical precedent suggests that leaderboard leadership in LLM benchmarks has shifted considerably over eighteen-month windows. OpenAI held sustained dominance through 2023–2024, but Anthropic's Claude series and open-source alternatives gained ground on specific evaluation suites. The current 89% implied probability reflects confidence that the incumbent leader (likely OpenAI) will retain its position, though this assumes no major architectural breakthroughs or benchmark methodology changes occur before the settlement date. Comparable prediction markets on model rankings have typically resolved within the top three contenders, with surprise outcomes rare when the probability exceeds 80%.

Traders should monitor scheduled model releases from OpenAI, Anthropic, Google DeepMind, and xAI throughout 2025 and early 2026. Any announcement of substantially larger parameter counts, novel training techniques, or expanded reasoning capabilities could shift the leaderboard. The Arena.ai platform itself occasionally updates its evaluation methodology; changes to the test set or scoring rubric could alter rankings independent of model improvements. Regulatory developments affecting model deployment in key markets may also influence which companies prioritise leaderboard performance versus other objectives.

Methodology

This page is a comparison snapshot: one live quote, four reference venues with their key attributes, and a single execution path — every trade button routes to PolyGram, which mirrors the Polymarket order book directly.

Resolution & payout

At resolution the UMA oracle takes over: a proposer posts the outcome with a bond, any token holder can dispute within two hours. Without dispute the result is accepted and the smart contract distributes USDC instantly.

On Kalshi (CFTC-regulated) resolution runs through their in-house clearing engine in USD. Betfair Exchange settles after match end in the account's local currency. Manifold pays no cash — only its in-platform "mana" currency.

FAQ

Where can I trade this market with the lowest fees?

Polymarket is geo-blocked in the US/UK/EU. The easiest 0%-fee broker into the same order book is PolyGram. Kalshi charges up to 7% per trade; Betfair Exchange takes 2-5% commission on net winnings.

What's the difference between YES and NO shares?

A YES share pays $1.00 if the event happens, $0 otherwise. A NO share pays $1.00 if the event doesn't happen. The market price between 0¢ and 100¢ is the implied probability.

What does Polymarket cost to trade?

Polymarket itself charges 0% — the only cost is the Polygon network fee, typically under $0.01 per transaction. Off-chain venues like Kalshi or Betfair charge 2-7% commission.

How fast are USDC deposits?

Polygon credits deposits after 12 confirmations — usually under 30 seconds. Withdrawals follow the same path and land back in your wallet within minutes.

How reliable are the quoted odds?

The YES/NO percentages are the live mid-prices of the Polymarket order book. On deep markets they move every few seconds; on thinner ones you'll see short plateaus.