Google takes the Gemini Interactions API to general availability, replacing generateContent as the default interface

Google has moved its Interactions API to general availability, positioning it as the new default way to build with Gemini models and agents in place of the long-standing generateContent endpoint.

What's new

According to Google's own developer documentation, "The Interactions API is our new interface and the most straightforward way to build with Gemini models and agents," and as of June 2026, "it is Generally Available and the recommended interface for all new projects."

The core change is server-side conversation state. Instead of developers resending full chat history on every call, the API introduces a previous_interaction_id parameter that lets Gemini reconstruct context automatically. Google's docs note this also improves caching: using previous_interaction_id allows "the system to more easily utilize implicit caching for the conversation history," cutting token costs on multi-turn conversations.

Each turn is represented by a new Interaction resource — a chronological record of "execution steps," including model reasoning, tool calls, tool results, and final outputs. That gives developers visibility into agent behavior for debugging and UI rendering that the older endpoint didn't expose directly.

The API also unifies access to Gemini models and Google's specialized agents — including Deep Research (standard and max) and Antigravity — behind a single endpoint, along with a background=true flag for long-running tasks that don't block on a response.

Supported models span Gemini 3.5 Flash, Gemini 3.1 Pro Preview, the Gemini 2.5 Pro/Flash line, and Gemma 4 open models, plus multimodal variants like Gemini 3.1 Flash TTS Preview and Gemini 3 Pro Image. Developers need google-genai 2.3.0+ (Python) or @google/genai 2.3.0+ (JavaScript) to use it.

By default, interactions are stored server-side for 55 days on paid tiers and one day on the free tier; store=false opts out but disables background execution and previous_interaction_id continuity. Some capabilities — video metadata configuration, Batch API integration, automatic function calling in Python, explicit caching — remain unavailable at GA, and remote MCP support for Gemini 3 is listed as "coming soon."

Context

generateContent has been Gemini's primary API surface since launch, but it pushed conversation-history and tool-orchestration bookkeeping onto the client. As agentic use cases grew — multi-step tool calls, long-running background tasks, mixed model/agent workflows — that bookkeeping became a bigger tax on every integration. The Interactions API is Google's answer, echoing a broader industry shift (OpenAI's Responses API, Anthropic's session-based agent APIs) toward server-managed state as the default for agent-building.

Why it matters

General availability, not preview, signals Google now expects new Gemini integrations to default to this interface rather than generateContent. For teams building agents, the built-in execution-step trace and automatic history management remove a real chunk of custom infrastructure that's currently duplicated across products. The token-cost win from better caching is also concrete rather than incidental — for high-volume multi-turn applications, that directly affects unit economics. The gaps that remain (explicit caching, Python auto function-calling, MCP support) suggest this is a GA foundation Google will keep building on, not a finished, feature-complete replacement yet.