Details
- Google has moved its Gemini Interactions API to general availability, making it the primary interface for both model inference and agent orchestration, with a single unified endpoint covering server-side state, background execution, tool use, and multimodal generation.
- The launch centers on Gemini models and managed agents, including the default Antigravity coding agent, with access via Google AI Studio, Python and JavaScript SDKs, and partner platforms LiteLLM, Eigent, and Agno.
- New capabilities include Managed Agents running in remote Linux sandboxes, async background execution via a background flag, richer tool combinations that blend Google Search and Google Maps with custom functions, and multimodal output including images, charts, music, and multi-speaker TTS.
- The Interactions API introduces a simplified, step-based schema replacing role messages, Deep Research variants optimized for speed or depth, Flex and Priority pricing tiers (with up to 50% cost reduction on Flex), and 55-day interaction retention on paid plans, while the legacy generateContent API remains supported but is no longer the default.
- Google is positioning the API as the foundation of an agent-first ecosystem, making it the default in Google AI Studio and Gemini API, publishing a migration guide, and shipping the gemini-interactions-api skill so coding agents can automatically apply best practices such as streaming, structured outputs, tool calling, and Deep Research workflows.
Impact
By unifying model and agent workflows behind a single, stateful endpoint, Google is steering Gemini usage toward long-running, agentic applications and shifting new development away from legacy generateContent patterns. Over the next 12–24 months, this design is likely to accelerate adoption of managed agents, deepen reliance on Google-native tools and media models, and influence competing platforms to further converge around agent-centric, server-side workflow APIs.
