LMArena November Check: Google's Gemini Leads Claude as Traders Flip to 62% for Google
Four days ago, the November check on the best AI model of 2026 was a coin flip. On 28 October, Polymarket priced Google at 50.0% and Anthropic at 45.0% to own the top-ranked model when the LMArena Text Arena (Overall) leaderboard is checked on 30 November 2026 at 12:00 PM ET, with style control off. Today that market has moved, and it has moved by more than the usual noise: Google is now priced at 62%, Anthropic at 32%, and OpenAI at just 2.4%. Google has gained about twelve points, Anthropic has lost about thirteen, and a market that was undecided now has a clear favourite.
The leaderboard itself explains most of the move, and it is worth looking at the actual numbers rather than the odds alone. On the board I checked today, Google holds the top two places: Gemini 4 Carbon (High) at 1540, and Gemini 4 Argon (High) at 1529. Anthropic's best model, Claude Opus 5.5 (High), sits third at 1516, with Claude Fable 5.1 (Max) just behind at 1514. The gap between Argon and Opus 5.5 is thirteen points, larger than either model's stated uncertainty, and Google's lead is not carried by a single model. That depth matters, because the market resolves on whichever company owns the single top-ranked model, and Google currently owns the top two.
There is a caution that belongs in the same paragraph as the lead. Carbon's 1540 comes from only 4,403 votes, far fewer than Argon's 11,889 or Opus 5.5's 11,415, so its score is the most likely to move as more votes accumulate. If Carbon fades, the contest narrows to Argon against Opus 5.5, and thirteen points is a lead, not a lock. Anthropic also has more than one model within striking distance, and the monthly crown has changed hands before: Anthropic won the June and July checks, Google won August. A month is a long time in this race, and a single new release can redraw the board.
What makes this flip interesting to me is that it runs against the week's other Google story. Alphabet's shares fell about 13% on 29 October, its largest single-day percentage drop on record, and closed Friday at $283.82 after losing roughly $500 billion in market value in a day. The stock market is charging Google for the cost of the AI buildout. The model market is pricing something different: not the company's balance sheet, but the quality of its models as judged, blind, by thousands of human voters. Those two verdicts can both be true at once, and the gap between them is the story of this autumn in AI. One market says the buildout is expensive; the other says, for now, it is working.
I am not copying the market's price into my own number, but I am close to it, for a reason I can state plainly. My prediction is that Google will own the top-ranked model on the LMArena Text Arena (Overall) leaderboard, style control off and excluding models marked AutoEval, when it is checked on 30 November 2026 at 12:00 PM ET, and I put that at 60%, against the market's 62%. The board lead is real and two models deep, and the December market has moved the same way, to 52.5% for Google against 36.5% for Anthropic. The discount from 62% to 60% is small because the case against is real too: Carbon's thin vote count, Anthropic's depth, and a full month in which a new model could arrive. If Anthropic retakes the top spot by the November check, that will be the informative miss, and it will say something important about how quickly a lead built on one board can be taken back.