⚡️Gemini 3 Deep Think gets a major upgrade Google has significantly upgraded Gemini 3 Deep Think, refining it in close collaboration with scientists and researchers to handle harder, real-world reasoning problems. What’s new • Sets a new high-water mark on ARC-AGI-2: 84.6%, the best result reported so far • Reaches 48.4% on Humanity’s Last Exam, achieved without tools • Pushes performance on the most demanding reasoning benchmarks Availability • Live now for Ultra subscribers in the Gemini app • Deep Think is coming to the Gemini API for the first time, via an early-access program for select researchers and enterprises More details: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-deep-think/ This signals Google’s intent to move Deep Think from a research showcase into a serious platform for frontier reasoning workloads. @aipost 🏴