Develop with Gemini API

Gemini API Development Skill guides application work with Gemini API hosted models, including Gemini and Gemma 4. Its scope includes text, images, audio, and video, alongside function calling, structured outputs, and model specification lookup. It pairs SDK selection with executable quick-start patterns and a documentation workflow for checking API details before implementation changes.

The model-selection guidance distinguishes fast, balanced multimodal use from complex reasoning, coding, research, lightweight high-frequency tasks, and image generation or editing. Named options include gemini-3.5-flash, gemini-3.1-pro-preview, gemini-3.1-flash-lite-preview, gemini-2.5-pro, and gemini-2.5-flash. Image-oriented entries include gemini-3-pro-image-preview, gemini-3.1-flash-image-preview, and gemini-3.1-flash-lite-image-preview, with the Nano Banana Pro, Nano Banana 2, and Nano Banana 2 Lite names. Gemma entries distinguish gemma-4-31b-it, a dense model, from gemma-4-26b-a4b-it, a mixture-of-experts model.

The workflow treats documentation lookup as an implementation step. When the Google MCP search_docs tool is available, queries and returned documentation become the source for API details, and manual URL fetching is avoided. Otherwise, the official documentation index is used to discover pages and retrieve focused Markdown documentation. Lookup topics include text generation, function calling, structured outputs, image generation, image understanding, embeddings, and SDK migration.

Use it when building applications against the Gemini API, handling multimodal content, adding callable functions, producing structured output, or selecting API and SDK details for an implementation. Its upstream product and API scope determine when it applies.

The selection rules exclude gemini-2.0-* and gemini-1.5-* and the legacy google-generativeai Python and @google/generative-ai JavaScript SDKs. Before changing an application, verify commands, API behavior, pricing, quotas, credentials, and deployment effects against official documentation. Generated examples still require environment-specific tests and security review; destructive or costly actions require user approval.

The quick-start pattern takes prompt content and a model identifier, creates a client, requests content generation, and prints the response text. The Python example is:

from google import genai

client = genai.Client()
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents="Explain quantum computing"
)
print(response.text)

The same API workflow supports JavaScript/TypeScript, Go, and Java. Documentation queries supply the API-specific details needed to extend these patterns to the supported content and output modes.

SDK guidance names google-genai for Python, @google/genai for JavaScript and TypeScript, google.golang.org/genai for Go, and com.google.genai:google-genai for Java. Java setup includes Maven and Gradle dependency forms with version lookup. For real-time bidirectional audio, video, and text streaming, the workflow points to the separate Gemini Live API development skill, covering WebSockets, voice activity detection, native audio, function calling, session management, and ephemeral tokens. The skill carries an Apache-2.0 license.