DeepSeek AI Models Now Available on Google Cloud's Gemini Enterprise Agent Platform

Google Cloud has significantly expanded the capabilities of its Gemini Enterprise Agent Platform by integrating a suite of advanced DeepSeek AI models, making them readily available to enterprise customers. This strategic collaboration, announced on May 29, 2026, empowers developers with greater flexibility and a broader selection of powerful AI tools for building and deploying sophisticated applications. The newly available DeepSeek models include DeepSeek-V3.1, DeepSeek-V3.2, DeepSeek R1 (0528), and DeepSeek-OCR, each bringing distinct and cutting-edge functionalities to the Google Cloud ecosystem. DeepSeek-V3.1 is highlighted for its innovative hybrid thinking modes and enhanced tool calling capabilities, offering a more nuanced and interactive AI experience. Building upon this, DeepSeek-V3.2 represents a leap forward in harmonizing high computational efficiency with superior reasoning and agent performance. This model's advancements are rooted in three key technical breakthroughs: DeepSeek Sparse Attention (DSA), a scalable reinforcement learning framework, and a large-scale agentic task synthesis pipeline. These innovations enable V3.2 to tackle complex problems with remarkable efficacy and speed. For tasks demanding robust reasoning and inference, the DeepSeek R1 (0528) model is now accessible, boasting significantly improved depth of reasoning and inference capabilities compared to its predecessors. It excels across a wide spectrum of applications, including creative writing, general question answering, text editing, and summarization. Complementing these language and reasoning models is DeepSeek-OCR, a comprehensive Optical Character Recognition solution designed to analyze and understand intricate documents. DeepSeek-OCR is particularly adept at handling challenging OCR scenarios, such as recognizing mathematical formulas and processing text that is curved, rotated, or overlapping, making it invaluable for industries dealing with diverse document types. Enterprise users can access these DeepSeek models through the Gemini Enterprise Agent Platform in two primary ways: as fully managed APIs or as self-deployed models. For managed models, developers can send requests directly to the Agent Platform API endpoint using specific model names like `deepseek-v3.2-maas` or `deepseek-ocr-maas`, eliminating the need to provision or manage underlying infrastructure. The platform also supports streaming responses, utilizing server-sent events (SSE) to incrementally deliver output, thereby reducing the end-user's perception of latency. For those preferring self-deployment or requiring specific commercial use licenses, the models can be enabled via the Model Garden console. It is important to note, as specified in the documentation, that DeepSeek models are not Google products, and their availability is subject to the terms for "Separate Offerings" within Google Cloud's AI/ML Services section and relevant model cards. This integration underscores Google Cloud's commitment to fostering an open and versatile AI environment, allowing customers to leverage best-in-class models from various providers to meet their specific business needs.