2 min read
Add as a preferred source on Google

Gemini 4 Carbon Scores 69.1% to Lead HLE-Diamond Ranking

The model leads both reasoning and knowledge sections. A separate Index v5 ranking was published before Google announced Carbon.

A clear glass diamond rests on unmarked question sheets / TokenPost.ai
A clear glass diamond rests on unmarked question sheets / TokenPost.ai

Gemini 4 Carbon (High) tops the current HLE-Diamond ranking with a 69.1% score, ahead of Claude Opus 5.5 (Max, Default Fallback) at 67.9%, giving Google’s model the highest listed result on this academic benchmark.

Claude Opus 5.5 (Xhigh, Default Fallback) scored 64.0%. Carbon also leads the ranking’s two sections, with 79.6% on reasoning and 58.6% on knowledge. The Max, Default Fallback Claude configuration scored 78.0% on reasoning and 57.8% on knowledge, leaving the models separated by 1.6 percentage points in reasoning and 0.8 points in knowledge.

HLE-Diamond contains 1,000 questions, split evenly between reasoning and knowledge, and 272 questions include images. The evaluation runs without tools and allows one answer per question. GPT-5.6 Luna (medium) judges the responses. The test focuses on closed-ended academic questions, so a leading score measures performance on that task set rather than a model’s ability to conduct open-ended research.

HLE-Diamond is a refined subset of Humanity’s Last Exam. The Center for AI Safety and Scale AI released it Sept. 22 after a year of refining questions. The organizations cautioned that high accuracy alone would not demonstrate autonomous research ability or artificial general intelligence.

Index v5, published Tuesday, Oct. 27, offers a separate, broader comparison. It ranked Claude Opus 5.5 first overall at 58, followed by GPT-6 Astra at 57 and GPT-6.1 Sol at 55. HLE-Diamond accounts for 10% of the index’s General category, which makes up 25% of the full index. Google announced Gemini 4 Carbon Wednesday, Oct. 28, one day after the index release, so those overall rankings do not include Carbon.

The HLE-Diamond result is a single-benchmark comparison. Its ranking focuses on a fixed question set, while Index v5 combines several categories. The two results therefore measure different ranges of model performance.

Google’s staged rollout, announced Wednesday, Oct. 28, starts with some cyber defenders and is set to expand to trusted partners. The company gave no date for general access.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…