Gemini 3 Pro for Developers: Getting Started with API and Use Cases

Introducing Gemini 3 Pro

• 🚀 Gemini 3 Pro is presented as Google's most intelligent model to date, excelling in science, math, multimodal reasoning, agentic tasks, and coding.
• 💡 The model is now available in preview for developers, with this video guiding users on how to start building and creating with it.

Data Analysis with Code Execution

• 📊 In Google AI Studio, users can test Gemini 3 Pro by uploading data, such as fruit stand sales transactions, and asking the model to analyze it.
• 💻 By enabling code execution, the model can process the data and return results formatted as an HTML file, including insights on sales, revenue, and top-performing items.

Getting Started with the Gemini API

• 🔑 Developers can use the Gemini API via SDKs (Python, Go, JavaScript) after creating an API key in Google AI Studio.
• 💰 Billing setup is required for API usage, and API keys should be securely managed, for example, as Colab secrets or environment variables, not hardcoded.
• ⚙️ The GenAI SDK needs to be imported and the client set up, with instructions provided for different development environments.

Advanced Model Parameters and Multimodal Capabilities

• 🤔 The API includes a thinking level parameter (high or low) to control the depth of reasoning, with 'low' suitable for straightforward prompts and 'high' for complex tasks.
• 🖼️ Gemini 3 Pro demonstrates multimodal reasoning by analyzing an image and a video simultaneously to find a specific scene, showcasing its ability to process different media types.
• 🔊 The model can identify a scene in a video based on a reference image, even with varying media resolutions to optimize token usage.

Integrating Built-in Tools

• 🌐 Gemini 3 Pro can leverage Google Search to gather information, such as identifying colors and hex codes for the 'dark academia' aesthetic, and then outputting this as runnable Python code.
• 🛠️ It can also combine tools like code execution and Google Search to perform complex calculations, such as determining the Flesch-Kincaid grade level of a literary work by extracting text and executing code.
• 🧠 For tasks requiring extensive research and computation, setting the thinking level to 'high' is recommended.