HLE-Diamond

Every result

• Gemini 4 Hydrogen58.8%Reported by the model's maker (Artificial Analysis leaderboard, as cited by Google), Nov 12, 2026

  Google reports 58.8% for Gemini 4 Hydrogen on HLE-Diamond, citing the Artificial Analysis leaderboard.

  Tasks: Diamond subset of Humanity's Last Exam, as labeled by Google; subset definition not checked.

  Setting: Google says its results use the Gemini API at the highest thinking setting (xhigh for Hydrogen) unless noted; no numeric budget is stated.

  Limits: Provider-reported: Google publishes the result for its own model; it has not been independently reproduced here. Google says this score comes from a third-party leaderboard; that leaderboard was not opened here, so its run date, harness and settings are unverified. A subset score, kept apart from full Humanity's Last Exam results. Google says the scores are as of November 2026 for the model it calls Gemini 4 Hydrogen; equivalence to the build now served on every surface is not established. Results under the same benchmark name can be shown together as reported scores, but unmatched settings and missing methods rule out a controlled ranking.

  Google DeepMind · Google DeepMind · Google DeepMind · Google DeepMind · Google

• Claude Sonnet 5.553.3%Independent evaluation (Artificial Analysis leaderboard, as cited by Google), Nov 12, 2026

  Google's Gemini 4 Hydrogen comparison table lists 53.3% for Claude Sonnet 5.5 on HLE-Diamond, citing the Artificial Analysis leaderboard.

  Tasks: Diamond subset of Humanity's Last Exam, as labeled by Google and Artificial Analysis; subset definition not checked.

  Setting: Google says it reports the maximum thinking setting available for Sonnet 5.5, or the best available result when that is not reported; no numeric budget is stated.

  Limits: Third-party report: Google, a competing model maker, published this figure for Anthropic's model in its own comparison table. Google reports this result for another maker's model in its Gemini 4 Hydrogen comparison table; it was not checked against that maker's or the leaderboard's own page. Results under the same benchmark name can be shown together as reported scores, but unmatched settings and missing methods rule out a controlled ranking.

  Google DeepMind · Google DeepMind · Google DeepMind · Google DeepMind