Google's fastest Gemini model delivers real latency gains for transcription and agentic workflows, but significant price hikes over 2.5 Flas