Google DeepMind released Gemini 3.5 Flash with the "Dynamic Thinking" feature, which lets the model dynamically decide how much reasoning to apply per query. Simple queries get fast responses; complex queries get deep reasoning. The "impossible triangle" of speed, cost, and quality is broken by dynamic allocation.