Three services covering the AI surface of Google Cloud — from the umbrella ML platform to the frontier LLM family to the specialized pre-trained APIs that still earn their place.
3 services
AI on Google Cloud changes faster than the rest of the platform — in 2026 the Vertex AI platform itself was rebranded as the Gemini Enterprise Agent Platform (this chapter keeps the Vertex name; the capabilities carry over). The decisions in this chapter — Vertex AI as platform versus lighter tooling, Gemini through Vertex versus direct, Gemini versus specialized pre-trained APIs — are less about absolute right answers and more about matching the tool to the workload's scale, governance needs, and cost profile. The shape of the choices is durable even as specific model versions and API surfaces evolve.
Vertex AIThe platform
End-to-end ML. Training, tuning, pipelines, model registry, endpoints, monitoring. Reach for it when you build, govern, or serve models at scale.
Gemini APIFrontier model
General reasoning. Text, vision, and multimodal generation. Use the direct API (Google AI Studio) for prototyping; run it on Vertex AI when you need IAM, governance, and regional controls.
Pre-trained APIstask-specific
Narrow, high-volume tasks. Vision, Speech, Translation, Document AI — managed APIs that are faster to adopt for a narrow task than building a custom model (still need quota, quality, and cost planning).
Vertex AIThe platform
Gemini APIFrontier model
Pre-trained APIstask-specific