HeadsUpAI

Google Launches Deep Research API and Moves Multimodal Embeddings to Production

Gemini
AI Economics
AI Agent
Multimodal
Performance
RAG

Google Launches Deep Research API and Moves Multimodal Embeddings to Production
Google moved Gemini Embeddings 2 to general availability, transitioning its first natively multimodal embedding model (a system that maps text and media into a single vector space) into a production-ready state. The update also introduces the Interactions API, providing programmatic access to the Deep Research and Deep Research Max autonomous agents.

This shift signals a move toward enterprise-grade reliability for multimodal RAG (grounding AI responses in external data) and agentic workflows. By implementing system-wide spending caps, Google is addressing the financial risks of autonomous agents. This follows the release of multimodal search applications and a universal design language.

You can now use Google AI Studio with increased limits through AI Pro or Ultra subscriptions. The environment now supports multi-chat contexts and web search grounding for retrieving live documentation. For consistency, the Stitch tool allows you to transfer design rules using the open-source DESIGN.md format.

Read the full update →

Frequently asked questions

How do I access Google AI Studio with a Google One subscription?
Google AI Studio is now included with AI Pro and Ultra plans, allowing subscribers to access the prototyping environment without needing a separate API key. This integration provides increased usage limits and premium features directly within the browser. Users can start building and testing Gemini models immediately using their existing Google subscription credentials.
What is the Google Deep Research API?
The Deep Research API allows developers to integrate Google's autonomous research agents, Deep Research and Deep Research Max, into their own applications. These agents are designed to perform complex, multi-step research tasks independently. Developers can access these capabilities through the Interactions API to build workflows that require deep information gathering and synthesis without constant human direction.
What are Gemini API billing account spending caps?
Gemini API billing account spending caps are a new system-wide safety feature designed to prevent unexpected overage charges. By setting a maximum limit on potential usage overrun, developers can ensure their projects do not exceed a defined budget. This is particularly important for autonomous agents that might otherwise consume significant resources during long-running or iterative tasks.
What is Gemini Embedding 2?
Gemini Embedding 2 is Google's first natively multimodal embedding model, now generally available for production use. It converts text, images, audio, video, and documents into a single vector space, enabling advanced semantic search and retrieval across different types of data. This allows developers to build more accurate retrieval-augmented generation systems that understand context across multiple modalities.
How does web search work in Google AI Studio?
Web search support in Google AI Studio allows the built-in AI agent to browse the internet for up-to-date information. This capability ensures that the model's responses are grounded in the latest data, such as current API documentation or recent news. Users can trigger this feature within the chat interface to improve the accuracy and relevance of generated code or information.