HLE-Diamond Leaderboard: Only 13% of Models Top 50%, Gemini 4 Carbon Leads at 69%
ArtificialAnlys · x · 2026-10-30
Artificial Analysis launched the HLE-Diamond leaderboard, an independent evaluation of the refined 1,000-question version of Humanity's Last Exam (released by CAIS and Scale AI after a year of cleaning) as part of its Intelligence Index v5. Questions are closed-ended, split evenly between reasoning and knowledge, 272 with images; every question is run once without tools and graded with the official HLE prompts by GPT-5.6 Luna (medium), reporting pass@1.
Key takeaways:
- Only 17 model variants (13%) score above 50%: Gemini 4 Carbon (high) leads at 69%, Claude Opus 5.5 (max) at 68%
- Models are stronger at reasoning than knowledge: over 90% score higher on reasoning. Gemini 4 Carbon posts 80% reasoning / 59% knowledge; Opus 5.5 and GPT-6 Astra (max) both 78% reasoning, with Opus 5.5 ahead on knowledge 58% vs 50%
- Open-weight models trail by 30+ points: Kimi K3.1 Pro (weights unreleased) tops at 37%, MiniMax-M3.1-Flash 28%, Kimi K3 (max) 27%
Related event: Artificial Analysis Launches HLE-Diamond Benchmark(3 posts)→
More from Models
- DeepSeek V4.1 Flash tops open-model usage among research agents, index data shows — AkariAsai · 2026-10-30
- Dev claims Gemini Argon makes 'researcher-level jumps' other models miss as 'carbon' looms — vitaliychiley · 2026-10-30
- Meta's Alexandr Wang touts Muse: 6.7x more concise than ChatGPT at answering the same question — alexandr_wang · 2026-10-30
- OpenAI's subsidy shift rewrites the score-vs-cost Pareto frontier, analysis finds — RexDouglass · 2026-10-30
- Phonic tops VAmoS benchmark for agentic speech-to-speech models — graceisford · 2026-10-30
- DeepSeek Frontend Leaks cyber-v3/v4 Keys; Still Hiring Attackers as V4 Excels at Security — teortaxesTex · 2026-10-30