Announcement
CritPt v2 launches with GPT-6 Astra (max) averaging 83.1% accuracy
The CritPt team says it changed 47 physics challenges after expert review, including problem-statement corrections and grader fixes.
TLDR
CritPt v2 is live on Artificial Analysis, which says models can write and run code in an offline sandbox. The CritPt team says it changed 47 challenges after four months of expert review and reports that GPT-6 Astra (max) averaged 83.1% accuracy over five runs. Artificial Analysis places GPT-6.1 Sol (max) second at 82%.
Combined views
—
7 Sources, first seen ago
