# Gemini Model Versions Supported in the Google Skills Repository: Complete 2025 Guide

> Explore the 9 Gemini model versions supported in the google/skills repository for text chat image generation live streaming and embeddings Discover the recommended gemini-3.6-flash model for your needs.

- Repository: [Google/skills](https://github.com/google/skills)
- Tags: deep-dive
- Published: 2026-08-15

---

**The Google Skills repository supports 9 Gemini model versions across 4 capability categories—text/chat, image generation, live realtime streaming, and embeddings—with `gemini-3.6-flash` as the default recommended model for most use cases.**

The official **Google Skills** repository maintains a curated, enterprise-grade list of Gemini model versions designed for the Gemini Enterprise Agent Platform. These models are grouped by capability and represent the only versions guaranteed to be stable, fully-featured, and supported by the Gen AI SDKs used throughout the codebase. This guide covers every supported model, its intended use case, and how to invoke it correctly.

## Supported Gemini Model Versions by Capability

The definitive model registry lives in [`skills/cloud/gemini-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-api/SKILL.md), which categorizes models by capability and specifies recommended defaults versus additional options available on explicit request.

### Text and Chat Models

| Model | Role | Context Window | Use Case |
|-------|------|---------------|----------|
| `gemini-3.6-flash` | **Recommended default** | ~1 million tokens | Fast, balanced, multimodal text and chat |
| `gemini-3.5-flash` | Additional | ~1 million tokens | Older flash model; use when explicitly requested |
| `gemini-3.5-flash-lite` | Additional | Standard | High-frequency, lightweight inference |
| `gemini-3.1-pro-preview` | Additional | ~1 million tokens | Complex reasoning, coding, research tasks |

The `gemini-3.6-flash` model serves as the primary workhorse for most text generation tasks. According to the source documentation, it offers optimal latency-quality tradeoffs for enterprise workloads. The `gemini-3.1-pro-preview` variant targets specialized scenarios requiring deeper reasoning capabilities.

### Image Generation Models

| Model | Quality Level | Internal Codename |
|-------|-------------|-------------------|
| `gemini-3-pro-image` | High | "Nano Banana Pro" |
| `gemini-3.1-flash-image` | Medium | "Nano Banana 2" |
| `gemini-3.1-flash-lite-image` | Fast, lower quality | "Nano Banana 2 Lite" |

All three image generation models respond to multimodal prompts combining text with reference images. The `gemini-3-pro-image` model produces the highest fidelity outputs for production visual content.

### Live Realtime API Model

| Model | Capability |
|-------|-----------|
| `gemini-live-2.5-flash-native-audio` | Streaming generation with native audio support |

This specialized model enables real-time conversational interfaces with low-latency audio output. It streams tokens progressively rather than waiting for complete generation.

### Embeddings Model

| Model | Purpose |
|-------|---------|
| `gemini-embedding-2` | Text embeddings for retrieval and semantic search |

The embeddings model generates dense vector representations optimized for RAG (Retrieval-Augmented Generation) pipelines and similarity-based search systems.

## Deprecated and Legacy Gemini Model Versions

The [`skills/cloud/gemini-interactions-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-interactions-api/SKILL.md) file explicitly lists deprecated model families that **must not be used** in new implementations:

- `gemini-2.5-*`
- `gemini-2.0-*`
- `gemini-1.5-*`
- `gemini-1.0-*`
- `gemini-pro`

These legacy families remain functional but will be removed from the platform. The Interactions API skill enforces runtime warnings when deprecated models are requested, though the underlying SDK does not block them.

## Using Gemini Models with the Gen AI SDK

All supported Gemini model versions are accessed through the **Google Gen AI SDK**, which operates in Enterprise mode with `GOOGLE_GENAI_USE_ENTERPRISE=true`. The SDK automatically routes requests based on model name and location settings.

### SDK Installation and Configuration

**Python:**

```bash
pip install google-genai

```

**TypeScript/JavaScript:**

```bash
npm install @google/genai

```

**Go:**

```bash
go get google.golang.org/genai

```

**Java:**

```xml
<dependency>
    <groupId>com.google.genai</groupId>
    <artifactId>google-genai</artifactId>
</dependency>

```

**Environment variables (all platforms):**

```bash
export GOOGLE_GENAI_USE_ENTERPRISE=true
export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_LOCATION=us-central1  # or "global" for automatic routing

```

### Code Examples for Each Supported Model

**Text generation with `gemini-3.6-flash` (Python):**

```python
from google import genai

client = genai.Client()
response = client.models.generate_content(
    model="gemini-3.6-flash",
    contents="Explain quantum computing in plain language."
)
print(response.text)

```

**Text generation with `gemini-3.6-flash` (TypeScript):**

```typescript
import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI();
const resp = await ai.models.generateContent({
  model: "gemini-3.6-flash",
  contents: "Explain quantum computing in plain language."
});
console.log(resp.text);

```

**Image generation with `gemini-3-pro-image`:**

```python
from google import genai

client = genai.Client()
resp = client.models.generate_content(
    model="gemini-3-pro-image",
    contents=[
        genai.Image.from_uri("gs://my-bucket/sample.jpg"),
        genai.Text("Create a stylized version of this photo.")
    ]
)
resp.image.save("stylized.png")

```

**Live realtime streaming with `gemini-live-2.5-flash-native-audio`:**

```python
from google import genai
import asyncio

client = genai.Client()
stream = client.live.create(
    model="gemini-live-2.5-flash-native-audio",
    audio=True
)

async def chat():
    await stream.send("Tell me a short story.")
    async for chunk in stream:
        print(chunk.text, end="")

asyncio.run(chat())

```

**Embeddings with `gemini-embedding-2`:**

```python
from google import genai

client = genai.Client()
emb = client.models.generate_content(
    model="gemini-embedding-2",
    contents="Searchable document about machine learning."
).embedding
print(emb[:10])

```

## Regional Deployment and Location Handling

By default, the Gen AI SDK uses `location="global"`, which enables Google to route requests to the nearest available region with capacity. For compliance or latency requirements, specify an explicit region:

```python
client = genai.Client(location="us-central1")

```

Available regions vary by model and are documented in the Gemini API skill configuration.

## Reference Implementation Files

| File Path | Purpose |
|-----------|---------|
| [`skills/cloud/gemini-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-api/SKILL.md) | Master model registry and SDK guidance |
| [`skills/cloud/gemini-interactions-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-interactions-api/SKILL.md) | Interactions API with deprecation warnings |
| [`skills/cloud/agent-platform-inference/scripts/gemini_genai_sdk.py`](https://github.com/google/skills/blob/main/skills/cloud/agent-platform-inference/scripts/gemini_genai_sdk.py) | Reference Gen AI SDK implementation |
| [`skills/cloud/agent-platform-inference/scripts/gemini_vertexai_sdk.py`](https://github.com/google/skills/blob/main/skills/cloud/agent-platform-inference/scripts/gemini_vertexai_sdk.py) | Legacy Vertex AI SDK example |
| [`skills/cloud/agent-platform-inference/scripts/gemini_openai_sdk.py`](https://github.com/google/skills/blob/main/skills/cloud/agent-platform-inference/scripts/gemini_openai_sdk.py) | OpenAI-compatible SDK example |

## Summary

- **9 Gemini model versions are officially supported** across text/chat, image generation, live realtime, and embeddings capabilities
- **`gemini-3.6-flash`** serves as the default recommended model for text and multimodal tasks
- **Legacy families (`gemini-2.5-*`, `gemini-2.0-*`, `gemini-1.5-*`, `gemini-1.0-*`) are deprecated** and trigger runtime warnings
- **All models require the Google Gen AI SDK** in Enterprise mode with `GOOGLE_GENAI_USE_ENTERPRISE=true`
- **Default `location="global"` enables automatic routing**, with explicit regions available for specialized requirements
- **Model enforcement logic resides in [`skills/cloud/gemini-interactions-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-interactions-api/SKILL.md)**, which implements the validation and warning system

## Frequently Asked Questions

### What is the default Gemini model version in the Google Skills repository?

The default recommended model is `gemini-3.6-flash`, a fast, balanced multimodal model with approximately 1 million tokens of context window. This is specified as the primary option for text and chat use cases in [`skills/cloud/gemini-api/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gemini-api/SKILL.md).

### Can I still use older Gemini models like `gemini-1.5-pro`?

Deprecated models including `gemini-1.5-*`, `gemini-2.0-*`, `gemini-2.5-*`, and `gemini-pro` remain technically functional but emit runtime warnings per the Interactions API skill. These families will be removed from the platform and should not be used in new code.

### How do I choose between image generation models?

Select `gemini-3-pro-image` for highest quality outputs ("Nano Banana Pro"), `gemini-3.1-flash-image` for balanced quality and speed ("Nano Banana 2"), or `gemini-3.1-flash-lite-image` for fastest generation with acceptable quality tradeoffs ("Nano Banana 2 Lite").

### What SDK is required to use these Gemini model versions?

All supported models require the **Google Gen AI SDK** (`google-genai` for Python, `@google/genai` for TypeScript/JavaScript, `google.golang.org/genai` for Go, `com.google.genai:google-genai` for Java). The SDK must operate in Enterprise mode with the environment variable `GOOGLE_GENAI_USE_ENTERPRISE=true` set.