We test DoubleDive's full pipeline on the hardest, newest exams students actually sit — tests released too recently to appear in any AI's training data — and publish the scores here.
Combined score across AIME 2026 I (15 problems) and JEE Advanced 2026 Math (34 problems).
American Invitational Mathematics Examination — the pathway to the USA Math Olympiad.
India's IIT entrance exam — widely considered one of the hardest math tests in the world.
Brand-new 2026 editions of the toughest exams American students sit.
A perfect SAT — Math and Reading & Writing, every question.
Perfect on the USA Physics Olympiad qualifier.
The USNCO Local exam.
A perfect Regents — every question.
On the 2026 Korean College Scholastic Ability Test benchmark, DoubleDive outscored every standalone frontier model — with perfect scores in Korean, English, Math, Physics, and Ethics.
All scores were measured internally in July 2026 on the newest available (2026) exam editions — released after every tested model's training cutoff, so none of the questions could appear in training data.
DoubleDive runs its full production pipeline: for each problem, an orchestrator routes to the best-suited model, has models cross-verify each other's work, and validates the result. Baselines are standalone single calls to each company's flagship model on the same inputs.
Korean CSAT scores are raw scores (원점수). SAT® (College Board), F=ma (AAPT), the U.S. National Chemistry Olympiad (American Chemical Society), and the Regents Examinations (NYSED) are trademarks of their respective owners, who do not endorse or sponsor DoubleDive.
Free credits on signup. No installation — start right in your browser.
Start Free