Gemini 3 Flash drops with Pro-grade reasoning, low latency

Gemini 3 Flash marks a paradigm shift where small models no longer compromise on reasoning depth for speed, matching larger models with 90.4% GPQA while running three times faster than previous Pro versions. The introduction of the thinking_level parameter and a competitive $0.50 per 1M input price point position it as the primary engine for high-volume agentic workflows and complex tool-use tasks.