Google Just Shipped Gemini 3.5 Flash. Here's What Developers Actually Need to Know.

Gemini 3.5 Flash challenges the long-standing speed-vs-intelligence tradeoff in AI models, delivering strong agentic performance at Flash-tier pricing. It introduces thought preservation across multi-turn conversations, a 1M token context window, and leads benchmarks like MCP Atlas at 83.6%. Teams migrating from Gemini 3 Flash Preview need to update model strings, replace thinking_budget with thinking_level, and test for quality changes due to the new default thinking level.