Gemini 3 is Google’s latest model family, emphasizing advanced reasoning and the ability to handle complex tasks. It supports agentic workflows, autonomous coding, and multimodal tasks. The main models include Gemini 3 Pro, designed for intricate tasks needing broad knowledge, and Gemini 3 Flash, which offers similar intelligence at a faster pace and lower cost. The Gemini 3 Pro Image model focuses on high-quality image generation. All these models are in preview, with a context window of up to 1 million tokens for text-based models and lower for image generation. A significant feature of Gemini 3 is the introduction of new parameters that give developers more control over the model’s performance. The thinking_level parameter allows users to adjust the depth of the model's reasoning, with options ranging from low for quick responses to high for detailed outputs. This is particularly useful for applications requiring rapid processing versus those needing careful analysis. Flash models include additional thinking levels, like minimal and medium, to cater to various task complexities. Gemini 3 also enhances multimodal capabilities, allowing developers to set media resolution for image or video inputs. Higher resolutions improve detail recognition but increase token usage and processing time. Recommended settings vary by media type, with specific guidance for images, PDFs, and video frames. For instance, high resolution is recommended for most image analysis tasks, while PDF document understanding typically saturates at medium resolution. Lastly, the temperature parameter is suggested to remain at a default of 1.0, balancing creativity and determinism in outputs.