Conceptual Reasoning Index
The Conceptual Reasoning Index (CRI) ranks models' abilities to reason about hard-to-verify topics like AI alignment and collective action problems under transformative AI.
Navigating the future requires hard-to-verify reasoning about important issues like AI alignment and collective action problems in the face of transformative AI. The Conceptual Reasoning Index (CRI) ranks models' abilities to reason about such topics. Leaderboard View Company Dataset 16/143 models · 2,435 data points #Model Overall (CRI)· 01 Claude Fable 5.1 74.7 02 GPT-6 Astra 73.8 03 Muse Spark 1.3 66.2 04 Gemini 3.8 Flash 65.0 05 GLM-5.3 63.4 06 Kimi K3 62.8 07 Grok 4.6 62.5 08 Qwen3.8-Max 59.4 09 DeepSeek V4 Pro (0813) 59.3 10 Inkling 52.6 11 Hy3…
saved by
related reading
- Introducing the Conceptual Reasoning Indexalignment.anthropic.com
- CAIS AI Dashboarddashboard.safe.ai
- SPAR Research Librarylibrary.sparai.org
- Learning to reason with LLMs | OpenAIopenai.com
- AI Model & API Providers Analysis | Artificial Analysisartificialanalysis.ai
- Coding vs thinking — Paradigm 3paradigm3.org
- Intelligenceintelligence.ai
- LMCA_dataset.pdfandrew.cmu.edu
- Robocurve — Real-world evaluations of physical AIrobocurve.org
- Kimi K3 Tech Blog: Open Frontier Intelligencekimi.com
- TERMINAL-BENCHtbench.ai
- FrontierSWEfrontierswe.com