Google favored for top AI model this month after Gemini 4 leads ranking

Google's Gemini 4 Carbon has been reported in first place on Arena's text leaderboard, and traders now see the company as the likely holder of the top spot when the ranking is checked at the end of November. Anthropic, the only other company given a real chance, is seen as unlikely to get there, and its odds have slipped from where they stood late last month. OpenAI is a long shot.

OfficeChai and GeoToolbox reported in early November that Gemini 4 Carbon (High) had moved ahead of Google's own Gemini 4 Argon on the leaderboard. That is the likely reason Anthropic's chances have faded, though the timing of the latest drop could not be matched exactly to those reports.

The path there was uneven. Anthropic's odds drifted lower through the first week of November, then jumped Monday afternoon while Google's fell. That was the day Amazon's cloud price list showed Claude Fable 5.2 and Mythos 5.2 before Anthropic announced them; the company later said Fable is available to all users and access to Mythos is restricted. The gain did not last. Anthropic's odds fell back overnight and through Tuesday morning to below where they started the month, while Google recovered all of its lost ground and a little more.

Arena's unadjusted text ranking showed Gemini 4 Carbon (High) first with 1,538 points, followed by Gemini 4 Argon at 1,530. Anthropic's Claude Opus 5.5 (High) was third at 1,513 and Claude Fable 5.1 (Max) fourth at 1,512. That leaves Anthropic's best-placed model behind two Google models. Carbon's score was marked preliminary. The page did not show a text ranking for the newly released Fable 5.2.

Traders are also cool on Anthropic for the following month: its chances of holding the top spot at the end of December are seen as unlikely and have eased since late October.

The question is settled by a single reading of Arena's overall text leaderboard, with style adjustments off, at midday on Nov. 30. The company whose model ranks first at that moment wins. Models carrying an automated-evaluation label are excluded, and a new release or a strong score on some other benchmark does not count on its own. New entries on the leaderboard, or changes to existing scores before the check, could still alter the result.