en.Wedoany.com Reported - Google recently announced the General Availability (GA) of the unified interface "Interactions API," which integrates the Gemini model and AI agents into a single entry point. Since its debut as a Beta version in December 2025, this API has become the most preferred method for developers, and this GA release marks its establishment as Google's foundational API for handling Gemini.

The Interactions API unifies model calls and agent execution into one endpoint. Passing a model ID enables inference, while passing an agent ID runs autonomous tasks; setting 'background=True' allows the server to process time-consuming tasks asynchronously. The API supports server-side state persistence, background execution, multi-tool composition, and multimodal generation, all accomplished with a single call.

This GA release includes numerous features highly requested by developers. The "Managed Agent" can launch a remote Linux sandbox with a single API call, enabling the agent to perform reasoning, execute code, browse the web, and process files. It provides a default "Antigravity" agent, and users can also create custom agents by defining their own instructions, skills, and data sources. The tool composition feature allows mixing built-in tools like Google Search and Maps with user-defined functions in the same request, returning both text and image results simultaneously.

The deep research feature is available in two versions: fast and deep, supporting collaborative planning, chart and infographic generation, as well as analysis based on images, PDFs, and audio. In terms of media generation, it offers Nano Banana 2 for image generation, Lyria 3 for music generation, and multi-speaker speech synthesis. The architecture has also been simplified from the original "Roles" structure to handle all actions as independent Steps.

In terms of cost, two tiers are available: Flex and Priority, allowing optimization based on cost or latency. Flex can reduce costs by 50%. The Interactions API is now the default interface for Google AI Studio, the Gemini API, and official documentation. The original generateContent API will continue to be supported, but the latest features for long-running models and agents will gradually be exclusive to the Interactions API. Stateful agent-based development is becoming the standard.