The following table lists the Gemini model capabilities with ML Processing support. ML Processing is not guaranteed for capabilities not explicitly listed in the following table. These capabilities apply to Gemini 2.5 Flash, Gemini 2.5 Pro, and their future versions.
For supported regional endpoints, see data residency.
| Capability | US and EU multi-regions | Country-level regions (excluding US) |
|---|---|---|
| Batch prediction | ||
| Chat completions | ||
| Code execution | ||
| Controllable parameters | ||
| Function calling | ||
| Security controls | ||
| Structured output | ||
| System instructions | ||
| Thinking budget | ||
| Single Zone PT (SZPT) | ||
| Context caching | ||
| Dynamic shared quota (DSQ) | ||
| Grounding with Vertex AI Search | ||
| PayGo | ||
| Provisioned Throughput | ||
| Supervised Fine Tuning | ||
| Thinking level | ||
| Thought summaries | ||
| Vertex AI RAG Engine |
For instructions on disabling unsupported capabilities in Vertex AI, see organization policy constraints.
What's next
- To quickly get started with Gemini, see Try a quickstart tutorial.
- To improve model performance by customizing models, see Tune models.
- To learn about regional data handling, see Data residency.
- To understand how to secure your generative AI applications, see Security controls.
- To discover and explore various models, see Explore models in Model
Garden.