Google Gemini 3 API updates for agents

Source

New Gemini API updates for Gemini 3

Why I saved it

This article is good because it shows how model APIs are becoming more like agent runtimes.

The important parts for me are thinking control, media resolution control, thought signatures, and structured outputs with grounding. These are product controls, not just model features.

My notes

• thinking_level gives a simple way to trade quality, cost, and latency.
• media_resolution matters because visual inputs can become expensive quickly.
• Thought signatures preserve reasoning continuity across tool calls.
• Grounding and URL context with structured output is very useful for extraction agents.

What I want to remember

Agent APIs need controls that map to product decisions: how hard to reason, how much image detail to inspect, when to ground, and what schema the result must follow.