DeepSeek models are available on Google Cloud Agent Platform

DeepSeek models are available on Gemini Enterprise Agent Platform as managed APIs and self-deployed models. Managed access lets developers send requests directly to the Agent Platform API endpoint without provisioning or managing infrastructure, while streaming responses can use server-sent events (SSE) to incrementally deliver output and reduce perceived end-user latency.

The available managed models include DeepSeek-OCR, DeepSeek-V3.2, DeepSeek-V3.1, and DeepSeek R1 (0528). DeepSeek-OCR is designed for complex document analysis, including mathematical formulas and curved, rotated, or overlapping text. DeepSeek-V3.2 focuses on computational efficiency, reasoning, and agent performance, while DeepSeek-V3.1 supports both thinking mode and non-thinking mode with improvements in tool calling and thinking efficiency. DeepSeek R1 (0528) improves reasoning depth and inference capabilities for tasks such as creative writing, question answering, editing, and summarization.

Google notes that DeepSeek R1 is not a Google product and is subject to separate offering terms and the relevant model card. For production-ready safety with DeepSeek R1 (0528), Google recommends using Model Armor to screen prompts and responses for security and safety risks. Regional availability varies, with DeepSeek-OCR, DeepSeek-V3.1, and DeepSeek R1 (0528) in us-central1, and DeepSeek-V3.2 available globally.