« Autopoiesis' new co-scientific AI surpasses models such as Grok 4 Heavy and Gemini 2.5 Pro (with a score of 92.4% on the GPQA Diamond benchmark). A major step forward toward scientific superintelligence »