Kagi, a search company, regularly quizzes AI models on a private set of tricky reasoning, coding and language questions that change over time, to see which ones actually think well.
Leaderboard
Top 3 right now
1stclaude-5-fable-thinking91.4
2ndclaude-5-fable88.8
3rdclaude-4-8-opus-thinking88.8
What it measures
Private, frequently changed set of reasoning, coding and instruction-following questions, reporting accuracy along with cost and speed.
How to read the numbers
Share of benchmark tasks answered correctly (%) — higher is better.
Note: Reasoning benchmark table, 116 tasks, last task list revision Feb 13, 2026; page has no explicit update date but lists models released after that. Order as published (by accuracy).