Artificial Analysis launches HLE-Diamond leaderboard, led by Gemini 4 Carbon

What happened

Artificial Analysis launched its HLE-Diamond leaderboard as part of v5 of the Artificial Analysis Intelligence Index, evaluating a refined 1,000-question version of Humanity's Last Exam. Each question runs once without tools, with GPT-5.6 Luna (medium) judging answers using the official HLE judge prompt to report pass@1. Gemini 4 Carbon (high) leads at 69%, followed by Claude Opus 5.5 (max) at 68%; only 17 model variants (~13%) exceed 50%, and over 90% perform better on reasoning than knowledge.

Read next

• More changes in today’s briefingMistral releases improved Mistral Large 4 open weights for coding and cybersecurity

The story so far

Artificial Analysis launched its HLE-Diamond leaderboard as part of version 5 of the Artificial Analysis Intelligence Index, with Gemini 4 Carbon (high) leading at 69% and Claude Opus 5.5 (max) following at 68%. The evaluation uses a refined, 1,000-question version of Humanity’s Last Exam. Each question is run once without tools, and GPT-5.6 Luna (medium) judges answers using the official HLE judge prompt to calculate pass@1. Artificial Analysis reported that only 17 model variants—about 13%—scored above 50%, and that more than 90% performed better on reasoning than on knowledge.