super-intelligence.compare
SOURCEDArena human preference · text / overallMethod & sources ↗Checked 12 Oct 2026

WHO’S RISING?

Each lab’s best closed-weight model against the best open-weight model, on the same Arena board.

Archive reconstruction: publisher-labelled history may contain later restatements. Not a backtest. Loading full history…

VIEW

Published board gaps; unsmoothed. Lines break when methodology changes.

Google leads · gemini 4 argon high · 42.2 above the baseline

-100-50050100Google42.2 ratingAnthropic23.2 ratingMeta-2.1 ratingAlibaba-8.3 ratingOpenAI-20.1 ratingxAI-43.6 rating8 Oct 20268 Oct 2026

Differences are meaningful within each board. Ratings across different pools and methodology versions are not an absolute capability trend.

8 Oct 2026
GoogleAnthropicXiaomiMetaAlibabamoonshotOpenAIzaiDeepSeekxAITencentMistral
Board published · 8 Oct 2026