New Gemini API updates for Gemini 3

Gemini 3 is available via API with updates for developers: new `thinking_level` for depth control, `media_resolution` for multimodal processing, and enforced `Thought Signatures` for agentic workflows, especially with function calling and image generation. It also introduces combining Google Search/URL Grounding with Structured Outputs and new usage-based pricing for Grounding. Best practices, like using default temperature, are advised for optimal results.

Google logo
Google
•
Shrestha Basu Mallick, Philipp Schmid
•4 min read•intermediate•
--
•View Original
GeminiJSONXML

Overview

The article discusses the latest updates to the Gemini API for Gemini 3, highlighting its advanced reasoning, coding capabilities, and multimodal understanding. Key features include simplified parameters for thinking control, granular control over multimodal vision processing, and the introduction of Thought Signatures to enhance function calling and image generation.

What You'll Learn

1

How to control the depth of reasoning in Gemini 3 using the thinking_level parameter

2

Why using Thought Signatures is essential for maintaining context in multi-step workflows

3

How to optimize media processing with the media_resolution parameter

4

When to use structured outputs for fetching live information from the web

Key Questions Answered

What is the purpose of the thinking_level parameter in Gemini 3?
The thinking_level parameter allows developers to control the maximum depth of the model's reasoning process before it generates a response. It can be set to 'high' for complex tasks or 'low' for applications sensitive to latency and cost, providing flexibility in how the model processes information.
How do Thought Signatures improve the performance of Gemini 3?
Thought Signatures are encrypted representations of the model's internal thought process that help maintain the chain of reasoning across API calls. This is crucial for complex workflows, ensuring that the rationale behind decisions is preserved, which enhances the overall quality of interactions with the model.
What changes were made to the pricing model for Grounding with Google Search?
The pricing model for Grounding with Google Search has shifted from a flat rate of US$35 per 1,000 prompts to a usage-based rate of US$14 per 1,000 search queries. This change aims to better support dynamic workflows and provide more granular cost management for users.
What are the best practices for using Gemini 3 Pro through APIs?
Best practices include keeping the temperature parameter at its default value of 1.0, maintaining consistency in prompt structure, and ensuring multimodal coherence by treating all input types equally. These practices help optimize the model's performance and output quality.

Key Statistics & Figures

New pricing for Grounding with Google Search
US$14 per 1,000 search queries
This new pricing model replaces the previous flat rate of US$35 per 1,000 prompts, allowing for more flexible cost management.

Technologies & Tools

Backend
Gemini API
Used for building applications with advanced reasoning and multimodal capabilities.

Key Actionable Insights

1
Utilize the thinking_level parameter to tailor the model's reasoning depth based on task complexity.
By adjusting the thinking_level, developers can optimize performance for specific applications, ensuring that the model's processing aligns with the needs of the task, whether it's a complex analysis or a simple data extraction.
2
Implement Thought Signatures in your API calls to enhance the quality of multi-step interactions.
Using Thought Signatures ensures that the model retains context and reasoning across interactions, which is particularly important for applications requiring detailed decision-making processes.
3
Adjust the media_resolution parameter to balance visual fidelity and token usage effectively.
By selecting the appropriate media resolution, developers can improve the model's ability to process images and videos while managing latency and costs, making it suitable for various use cases.

Common Pitfalls

1
Neglecting to use Thought Signatures can lead to degraded reasoning quality in the model's responses.
Without Thought Signatures, the model may lose context between API calls, resulting in less coherent and relevant answers, particularly in complex workflows.