Gemini API in Gemini platform
[!IMPORTANT] Gemini platform was formerly named Gemini Enterprise Agent Platform and Vertex AI.
Usage Attribution
MUST READ: Before executing any gcloud, REST API, or MCP command, you
MUST read and follow the instructions in
Usage Attribution.
Core Directives & Prerequisites
- Unified Gen AI SDK ONLY: Python
google-genai(pip install "google-genai>=2.33.0,<3"), TS/JS@google/genai(npm install @google/genai), Gogoogle.golang.org/genai(go get google.golang.org/genai), Javacom.google.genai:google-genai:2.6.0, C#Google.GenAI(dotnet add package Google.GenAI). - NEVER use deprecated SDKs:
google-cloud-aiplatform(vertexai),@google-cloud/vertexai,cloud.google.com/go/vertexai,google-cloud-vertexai, orgoogle-generativeai(Migration Guide). - Prerequisites: Active billing (Pricing),
roles/aiplatform.user, API enabled (gcloud config set project {project_id},gcloud services enable aiplatform.googleapis.com), ADC (gcloud auth application-default login). - Efficiency: Model IDs and SDK signatures here and in
references/are verified against the current SDK. Respond directly if answered below; otherwise read only the matching line range in Reference Routing without reading unrelated references or running speculative repo/web searches unless inspecting user files or unlisted APIs.
Authentication & Python Quick Start
Prefer environment variables (GCP Auth, Locations, Express Mode):
# ADC ('global' default; Live API and Tuning require regional, e.g. 'us-central1')
export GOOGLE_CLOUD_PROJECT='your-project-id'
export GOOGLE_CLOUD_LOCATION='global'
export GOOGLE_GENAI_USE_ENTERPRISE=true
# Express Mode API key (mutually exclusive with project/location — unset GOOGLE_CLOUD_PROJECT & GOOGLE_CLOUD_LOCATION):
# export GOOGLE_API_KEY='your-api-key' GOOGLE_GENAI_USE_ENTERPRISE=true
# Explicit Express Mode: enterprise=True+api_key (Py); enterprise:true+apiKey (TS/Java/C#); BackendEnterprise+APIKey (Go) — omit project/locationfrom google import genai
from google.genai import types # Only needed when passing optional config
# Zero-arg Client() reads env vars; explicit: genai.Client(enterprise=True, project="your-project-id", location="global")
client = genai.Client()
response = client.models.generate_content(
model="gemini-4-hydrogen",
contents="Explain quantum computing",
# Optional config:
# config=types.GenerateContentConfig(
# system_instruction="Be concise.",
# max_output_tokens=256,
# ),
)
print(response.text)Models
[!WARNING] NEVER recommend deprecated models:
gemini-2.5-*,gemini-2.0-*,gemini-1.5-*,gemini-1.0-*,gemini-pro*,gemini-pro-vision,text-embedding-004,multimodal-embedding-001,imagen-3.0*.
| Capability | Recommended Model(s) | Notes |
|---|---|---|
| General text, multimodal, coding, agents | gemini-4-hydrogen |
Default; 1M ctx, fast & balanced |
| Complex reasoning & enterprise SWE | gemini-4-neon |
1M ctx, long-term GA, broad regions |
| Frontier reasoning, security & research | gemini-4-carbon |
Max reasoning depth; 1M ctx, select regions |
| High-frequency / lightweight tasks | gemini-3.5-flash-lite |
Low latency & cost; 1M ctx |
| Text & multimodal embeddings | gemini-embedding-2 |
Shared space: text, image, PDF, audio, video |
| Image generation & editing | gemini-nano-banana-2.1, gemini-3.1-flash-lite-image |
Pro/high-res (2.1) or fast Lite |
| Video generation & editing (Omni) | gemini-omni-1.2-flash |
Up to 4K + audio via client.interactions.create |
| Video generation (Veo async) | veo-3.1-fast-generate-001 |
Default fast; also veo-3.1-lite-generate-001, veo-3.1-generate-001 |
| Live bidirectional audio/video | gemini-3.8-live |
client.aio.live.connect (regional endpoint) |
| Text-to-Speech (TTS) | gemini-3.8-flash-tts, gemini-3.8-flash-lite-tts |
Studio (132 langs) or low-latency (103 langs) |
| Music generation | lyria-3-pro-preview, lyria-3-clip-preview, lyria-002 |
<=184s / 30s (interactions.create); lyria-002 (GA :predict) |
Use ONLY if explicitly requested (or required by feature): gemini-4-argon, gemini-3.8-flash (explicit context caching), gemini-3.1-pro-preview, gemini-3.7-flash, gemini-3.6-flash, gemini-3.5-flash, gemini-3.1-flash-lite (SFT tuning), gemini-3-pro-image, gemini-3.1-flash-image.
Reference Routing
| Task | Reference & Line Range |
|---|---|
TS/JS (L12-L32), Go (L34-L67), Java (L69-L102), C# (L104-L123) |
references/sdks.md |
Chat (L11-L26), streaming (L28-L38), multimodal/media_resolution (L40-L89), async client.aio (L91-L111), count_tokens/usage (L113-L133) |
references/text_and_multimodal.md |
JSON schema (L10-L36), function calling (L38-L92), Search & Agent Search grounding (L94-L132), code exec & URL context (L134-L161) |
references/structured_and_tools.md |
Nano Banana images (L11-L53), Omni t2v (L55-L92), Omni i2v/ref/edit/async/multi-turn (L55-L155), Veo (L157-L187), TTS single/multi-speaker (L189-L268), Lyria (L270-L291) |
references/media_generation.md |
Text, image, and segmented video embeddings (EmbedContentConfig) |
references/embeddings.md |
Explicit caching (L10-L47), batch (L49-L74), thinking (L76-L108), MCP tools (L110-L138) |
references/advanced_features.md |
Safety filters (HarmCategory, HarmBlockThreshold) & ratings |
references/safety.md |
Bounding box detection ([y_min, x_min, y_max, x_max], 0-1000) |
references/bounding_box.md |
Live API WebSockets (client.aio.live.connect, PCM audio) |
references/live_api.md |
Supervised Fine-Tuning (SFT) & Preference Tuning (client.tunings.tune) |
references/model_tuning.md |
Official Docs & Endpoints
- Gemini platform Overview | REST Reference (global:
https://aiplatform.googleapis.com/v1beta1/projects/{PROJECT}/locations/global/publishers/google/models/{MODEL}:generateContent; regional:https://{LOCATION}-aiplatform.googleapis.com/...) | Python Samples - If
search_documents/get_document(Developer Knowledge MCP) are available, query them when unlisted API details are needed.