Gemini Developer API vs Gemini Enterprise Agent Platform (Vertex AI) — overview
Status: DOCUMENTED (overview only; the Enterprise Agent Platform was not called — no GCP credentials in this atlas). Sources: https://ai.google.dev/gemini-api/docs/migrate-to-cloud · https://ai.google.dev/gemini-api/docs/libraries · https://ai.google.dev/gemini-api/docs/zdr · https://ai.google.dev/gemini-api/docs/pricing · https://ai.google.dev/gemini-api/docs/deprecations (Veo GA note) · https://ai.google.dev/gemini-api/docs/model-tuning Last verified: 2026-09-18.
| Aspect | Gemini Developer API (this atlas) | Gemini Enterprise Agent Platform (Vertex AI) |
|---|---|---|
| Host | generativelanguage.googleapis.com/{v1beta,v1} (global) |
https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/publishers/google/models/{model}:generateContent (regional; global location available for some models) |
| Auth | API key (x-goog-api-key; auth keys bound to a service account), OAuth optional, ephemeral tokens for Live |
Google Cloud IAM: OAuth 2.0 / ADC / service accounts only; no API keys |
| Onboarding | AI Studio key in minutes; free tier | GCP project, billing, IAM roles |
| SDK | same google-genai / @google/genai / go / java; Client() |
same SDK with vertexai=True, project, location or GOOGLE_GENAI_USE_VERTEXAI=true |
| Model ids | gemini-3.8-flash, previews, -latest aliases, Veo/Lyria/Omni ids |
same base ids for Gemini (gemini-3.8-flash), Veo GA ids (veo-3.1-generate…), Model Garden third-party models (Claude, Llama…); preview availability and dates differ |
| API versions | v1beta (default) / v1 |
v1 / v1beta1 (different discovery surface) |
| Feature parity | Interactions API, Live API, Files API (48 h TTL), File Search stores, Managed agents/Antigravity, Deep Research, OpenAI-compat layer, ephemeral tokens | Vertex: RAG Engine, Agent Engine, supervised fine-tuning, provisioned throughput, model garden, evaluation, Cloud Storage/BigQuery inputs, VPC-SC, CMEK; Files API not available (use GCS URIs); OpenAI-compat exists but at a different endpoint |
| Data handling | Unpaid Services: content may be used to improve products; Paid: not used, 55-day abuse logging; ZDR not guaranteed (Search/Maps grounding 30-day storage) | enterprise DPA, zero data retention options, data residency, compliance certifications |
| Regions | availability by developer country (~190); single global endpoint | ~40 GCP regions with data residency; region-specific model availability |
| Pricing | pricing.md (free tier, Batch/Flex −50 %, Priority +80 %) | separate Cloud price list ("prices may differ"); provisioned throughput, committed-use discounts |
| Rate limits | project tiers Free/1/2/3, spend-based limits, AI Studio dashboard | per-region quotas (Cloud console), dynamic shared quota, provisioned throughput |
| Tuning | not offered (retired May 2025) | supervised tuning for Gemini |
| Migration notes (migrate-to-cloud.md) | delete unused API keys after moving (gcloud beta services api-keys undelete to recover) |
use service accounts; same or new project; models created in AI Studio must be retrained; supported regions differ |
Rule of thumb from Google: "Most developers should use the Gemini Developer API unless there is a need for specific enterprise controls." Both are reachable from one SDK, so code written against client.models.generate_content(...) / client.interactions.create(...) ports by changing the client constructor.