Google Cloud Generative AI Offerings and Capabilities
Two layers: models and platform
The frontier models are the Gemini family: Pro for complex reasoning, Flash for balanced price/performance/latency, and Flash-Lite for high-volume, low-cost tasks. Vertex AI is the enterprise platform around them: Model Garden (Gemini plus third-party and open models such as Claude and Llama), Vertex AI Studio for prompt design and prototyping, Agent Builder / Agentspace for agents, and Vertex AI Search for grounded retrieval. Specialized generators cover other modalities: Imagen (images), Veo (video), Chirp (speech-to-text), and Text-to-Speech.
Don't confuse the surfaces
Separate the developer platform from productivity experiences. Gemini for Google Cloud (Code Assist, Gemini in BigQuery, Gemini Cloud Assist) and Gemini in Workspace are assist experiences, not the Vertex AI API. Likewise, Model Garden is a catalog, not a single model.
Operations and protection
Evaluation, Model Monitoring, pipelines for MLOps, Model Armor for prompt-injection and harmful-content screening, and VPC Service Controls plus CMEK for the data perimeter.
Traps
- Assuming a named Gemini model requires an API outside Vertex AI.
- Assuming partner or open models are unavailable on Google Cloud.
- Defaulting to the largest model - cost and latency questions usually want Flash or Flash-Lite.
- Treating Gemini in Workspace as a build platform.
Rule of thumb
Match offering to job: prototype in Vertex AI Studio → build on Vertex AI with Model Garden → ground with Vertex AI Search → orchestrate with Agent Builder → shield with Model Armor. Start with Flash and escalate to Pro only when evaluations justify it.