Google has moved the Interactions API to general availability, establishing it as the primary interface for Gemini models and agents. Initially launched in a public beta in December 2025, the API provides a unified endpoint for model inference, autonomous agent execution, and multimodal generation with features like server-side state and background processing. Developers can now utilize Managed Agents, which automatically provision remote Linux sandboxes for reasoning, web browsing, and file management. The default Antigravity agent is included, while users can also define custom agents with specific data sources and skills.

The API introduces several key improvements, including a simplified schema that replaces legacy roles with typed steps such as thoughts, function calls, and model outputs. Background execution is now supported by setting a boolean flag to true. Enhanced tool capabilities allow for mixing built-in tools like Google Search and Maps with custom functions, supporting image returns. Deep Research upgrades offer two new agent versions for speed or depth, collaborative planning, and native chart generation. Media generation updates feature Nano Banana 2 for images, Lyria 3 for music, and multi-speaker text-to-speech.

For cost and performance, Google introduced Flex and Priority tiers, with the Flex tier offering a 50% cost reduction. Paid tier users benefit from 55-day interaction retention. While the legacy generateContent API remains supported, future frontier capabilities for agents will prioritize the Interactions API. The company has released the gemini-interactions-api Skill to help coding agents adopt these new development patterns. The API is accessible via Python and JavaScript SDKs, with support from partners including LiteLLM, Eigent, and Agno.

Google has moved the Interactions API to general availability, establishing it as the primary interface for Gemini models and agents. Initially launched in a public beta in December 2025, the API provides a unified endpoint for model inference, autonomous agent execution, and multimodal generation with features like server-side state and background processing. Developers can now utilize Managed Agents, which automatically provision remote Linux sandboxes for reasoning, web browsing, and file management. The default Antigravity agent is included, while users can also define custom agents with specific data sources and skills.

The API introduces several key improvements, including a simplified schema that replaces legacy roles with typed steps such as thoughts, function calls, and model outputs. Background execution is now supported by setting a boolean flag to true. Enhanced tool capabilities allow for mixing built-in tools like Google Search and Maps with custom functions, supporting image returns. Deep Research upgrades offer two new agent versions for speed or depth, collaborative planning, and native chart generation. Media generation updates feature Nano Banana 2 for images, Lyria 3 for music, and multi-speaker text-to-speech.

For cost and performance, Google introduced Flex and Priority tiers, with the Flex tier offering a 50% cost reduction. Paid tier users benefit from 55-day interaction retention. While the legacy generateContent API remains supported, future frontier capabilities for agents will prioritize the Interactions API. The company has released the gemini-interactions-api Skill to help coding agents adopt these new development patterns. The API is accessible via Python and JavaScript SDKs, with support from partners including LiteLLM, Eigent, and Agno.