New Gemini 3 API Updates Enhance AI Capabilities for Developers

The Gemini 3 model, DeepMind’s most advanced AI, is now accessible to developers via an updated API. Built to deliver superior reasoning, autonomous coding, multimodal understanding, and powerful agentic functions, the new updates aim to provide greater control and flexibility for various applications.

One significant addition is the new thinking_level parameter. This allows users to set the depth of the model’s internal reasoning process, balancing complexity with speed and cost. For complex tasks like strategic analysis or code review, set it to “high”; for simpler functions such as summarization, choose “low.”

The API now also offers refined control over media inputs with the media_resolution parameter. Users can adjust image, video, or document resolution levels—low, medium, or high—determining the trade-off between visual detail and resource consumption. Higher resolutions enable better recognition of fine details but increase processing time.

Another key update involves the introduction of thought signatures. These encrypted representations help track the model’s reasoning across multi-step workflows, ensuring consistency and clarity in decision-making. When used correctly, especially in function calling and image generation, these signatures improve accuracy and enable complex conversations. Omitting them may cause validation errors or degrade response quality.

Additionally, the API now supports integrating web-based tools like grounding with Google Search and URL context with structured outputs. This feature allows models to fetch live web data and extract information in structured JSON format, ideal for building dynamic, real-time data-driven applications.

Lastly, pricing for Grounding with Google Search has shifted from a flat rate to a usage-based fee, decreasing from $35 per 1,000 prompts to $14 per 1,000 search queries, making it more economical for extensive web integration.

These updates aim to enhance the versatility and performance of Gemini 3 API, encouraging innovative uses in areas such as complex reasoning, multimodal tasks, and web-enabled workflows.

In summary, the Gemini 3 API now offers more control over reasoning depth, media quality, and data integration—empowering developers to build smarter, more efficient AI-powered tools.

FAQs:

1. What is the purpose of the thinking_level parameter?
It controls the depth of the AI’s reasoning process, balancing complexity with performance for different tasks.

2. How does media_resolution affect media inputs?
It adjusts the level of detail in images or videos, affecting accuracy and processing time.

3. Why are thought signatures important?
They help maintain the reasoning chain across multiple interactions, improving decision consistency and quality.

4. Can I fetch live web data with the Gemini API?
Yes, the API now supports grounding with Google Search and URL context features for real-time data extraction.

5. How has the pricing changed for web search functionalities?
The cost has shifted from a flat $35 per 1,000 prompts to a usage-based rate of $14 per 1,000 queries.