The curved glass facade of Google's Googleplex headquarters in Mountain View, with the colorful Google logo on the building.
Image: Asoundd, CC BY-SA 4.0, via Wikimedia Commons

Best AI model

Google still leads bets for best AI model at end of November, but its edge shrinks

Google remains the favorite for the top spot on Arena's text leaderboard, but the reason for Friday's sharp drop is not yet clear.

Traders still see Google as the likeliest company to own the top-ranked AI model on Arena's text leaderboard at the end of November, but they cut that view sharply late Friday. Bets on Google fell from likely to a toss-up in a matter of minutes. Anthropic is the next closest contender, still unlikely to win but better placed than when the story began. OpenAI remains a long shot.

The reason for the drop is not yet clear. No news item or leaderboard change in the available information explains the timing, which came on Friday evening.

Until then the bets had been steady. Through Wednesday and Thursday, expectations for Google held in the likely range. They edged higher early Friday and stayed there through the afternoon before the sudden fall.

The backdrop favored Google. Arena's leaderboard, with style control off, showed Google's Gemini 4 Carbon (High) first at 1,540 in a table headed October 27. Google's Gemini 4 Argon (High) was second at 1,529, and Anthropic's Claude Opus 5.5 (High) was third at 1,516. Carbon carried a Preliminary label, not the AutoEval label that would exclude it. An earlier Arena snapshot, from October 15, had Argon in the lead.

Google announced Carbon on October 28, calling it its strongest model. It initially limited access to selected cyber defenders and trusted partners. That announcement came two days before the drop and is background, not a demonstrated cause of it.

The question is settled by Arena's Overall Text table, with adjustments set to None, checked at noon Eastern time on November 30. The company that owns the top-ranked model wins. A model marked AutoEval at that moment is excluded, but a Preliminary label alone does not disqualify one. Ties are broken by Arena score.

What happens next depends on the leaderboard itself. New model releases or reshuffled rankings before the check could change the outcome. Google says it plans to expand access to Carbon but has not given a date. Bets on Google finishing on top at the end of December remain a toss-up and did not move.

Sources

VersusApps

LMArena Leaderboard October 2026: Where the Top Text Scores Stand

↗

BenchLM

LLM Leaderboard & AI Model Benchmarks, October 2026

↗

NeuralTrust

Gemini 4 Argon vs GPT-6 Astra (2026)

↗

Wikipedia

Claude (AI)

↗

Move history

  • ▲ 21 ptsto 74%
  • ▼ 18 ptsto 52%

Odds reflect traders’ expectations, not facts, and can swing on thin trading.