Compared with the full Gemini 2.5 Pro, Gemini 2.5 Flash significantly reduces inference latency and computational cost while maintaining relatively high generation quality.