Tag
KINA (Knowledge Index of Noah's Ark) is an 899-item LLM benchmark spanning 261 fine-grained disciplines, introducing formal guarantees for disciplinary representativeness, incentive-aligned annotation via bonus-on-bar tournaments, and bootstrap ranking-stability reporting. Evaluating 42 models, top performers include Gemini-3.1-Pro-Preview (53.17%), Claude-Opus-4.6 (49.92%), and GPT-5.4 (48.55%), revealing a tiered rather than smooth leaderboard structure.