Multimodal AI with Gemini 2.5 and 3 models, supporting text, images, video, audio, PDFs, function calling, thinking mode, and real-time web grounding.

• Supports three deployment approaches: Node.js SDK (@google/genai), fetch-based REST API for edge runtimes, and chat helpers for multi-turn conversations
• Handles multimodal inputs (images, video, audio, PDFs) with 1,048,576 token context window; thinking mode enabled by default for enhanced reasoning quality
• Includes function calling with parallel execution, system instructions, streaming, and generation parameter control (temperature, top-p, stop sequences)
• Advanced features: context caching for cost optimization, Python code execution in sandbox, Google Search grounding for real-time information with citations
• Prevents 14 documented errors including SDK deprecation, safety setting thresholds, structured output JSON parsing, and December 2025 rate limit reductions