"Gemini 4 Argon: Google Leads the October AI Arena Race at 90% — After Flipping It in 20 Minutes"

On the evening of September 30, the prediction market for October's best AI model belonged to Anthropic. Traders priced the maker of Claude at 88.5 cents — an 88.5% chance of finishing October on top of the arena.ai text leaderboard. Then, within twenty minutes of 8 p.m. UTC, the board flipped. Google jumped from 10.3 cents to 62.5; Anthropic fell to 29.5. The trigger was a single announcement: Google's Gemini 4 Argon.

Argon arrived at the top of the leaderboard almost immediately, and it has not left. On the October 22 board it led at 1526 points, give or take seven, against 1516 for Claude Opus 5.5 and 1513 for Claude Fable 5.1. Polymarket's market for the October crown now prices Google at 89.9% and Anthropic at 8.0%, on $3.4 million of trading. The market resolves on one specific thing: whichever company owns the top-ranked model on the arena.ai Text Arena overall board, style control off, when it is checked at noon Eastern on October 31.

That specificity is the interesting part. This market does not measure which model is "best" in any general sense. It measures who sits on top of one table on one day. And the table is a strange one to crown a champion from: Argon is not generally available. Google has released it only to select cybersecurity partners through its Fairwind program, with wider access promised once safety guardrails are refined. The model the market calls the best in the world is, for now, a model almost nobody can use.

The lead is also thinner than the 90% price suggests. Ten points separate Argon from Opus 5.5, and the error bars overlap: 1526 ± 7 against 1516 ± 7. Anthropic, meanwhile, has depth that Google does not — nine of the top fourteen entries on the board are Claude models. And six days is a long time on a board like this one: the same market that now says 90% for Google said 88.5% for Anthropic a month ago, and it reversed in twenty minutes on a single product announcement. Meta's Muse Wonder, a new entry at 1501, is another reminder that a fresh release can scramble the board overnight.

So my prediction is Google, but not at the market's price. I put Google's chance of still holding the top spot at the October 31 check at 80%: a clear favourite, with a real chance of an upset. If Argon is still on top, the story will not be that Google built the best model in some absolute sense. It will be that the leaderboard crown — a title that has changed hands between OpenAI, Google, xAI and Anthropic over the last three years — spent October with a model most people cannot touch. That gap, between the crown and the thing it crowns, is the part I keep coming back to. A leaderboard can tell you who won the vote. It cannot tell you what the win is worth.