Run Gemma with the Gemini API¶

Models Gemma 4 Uses Gemini API

The Gemini API provides hosted access to Gemma as a programming API you can use in application development or prototyping. This API is a convenient alternative to setting up your own local instance of Gemma and web service to handle generative AI tasks.

Supported models¶

The Gemini API hosts the two largest Gemma 4 models:

Model	Identifier
Gemma 4 31B	`gemma-4-31b-it`
Gemma 4 26B A4B	`gemma-4-26b-a4b-it`

The smaller sizes are not hosted — run those yourself with Ollama, llama.cpp or LiteRT-LM. The selector on this page offers the two sizes the API serves, and every example below follows it.

PythonJavaScriptREST

from google import genai

client = genai.Client()

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    contents="Roses are red...",
)

print(response.text)

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

const ai = new GoogleGenAI();

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: "Roses are red...",
});
console.log(response.text);

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [{
    "parts":[{"text": "Roses are red..."}]
    }]
   }'

Get API Key

Important

You must obtain an API key to use the Gemini API, which you can get from the Google AI Studio application.

You can access the Gemini API on many platforms, such as mobile, web, and cloud services, and with multiple programming languages. For more information on Gemini API SDK packages, see the Gemini API SDK downloads page. For a general introduction to the Gemini API, see the Gemini API get started guide.

Thinking¶

Gemma 4 utilizes an internal "thinking process" that optimizes its multi-step reasoning, delivering superior performance in logically demanding domains such as algorithmic coding and advanced mathematical proofs.

While Gemma 4 strictly supports toggling this feature on or off, you can control it using the API by setting the thinking level to "high" for enabled or "minimal" for disabled.

The following example demonstrates how to activate the thinking process:

PythonJavaScriptREST

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    contents="What is the water formula?",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_level="high")
    ),
)

print(response.text)

import { GoogleGenAI, ThinkingLevel } from "@google/genai";

const ai = new GoogleGenAI();

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: "What is the water formula?",
  config: {
    thinkingConfig: {
      thinkingLevel: ThinkingLevel.HIGH,
    },
  },
});
console.log(response.text);

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [{
    "parts":[{"text": "What is the water formula?"}]
    }],
    "generationConfig": {
      "thinkingConfig": {
            "thinkingLevel": "high"
      }
    }
   }'

Learn more about Thinking:

Gemini API Thinking (General introduction)
Gemma Thinking (Gemma-specific capabilities)

Image Understanding¶

Gemma 4 models can process images, enabling many frontier developer use cases that would have historically required domain specific models.

The following example shows how to use Gemma Image inputs with the Gemini API:

PythonJavaScriptREST

from google import genai

client = genai.Client()

my_file = client.files.upload(file="path/to/sample.jpg")

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    contents=[my_file, "Caption this image."],
)

print(response.text)

import {
  GoogleGenAI,
  createUserContent,
  createPartFromUri,
} from "@google/genai";

const ai = new GoogleGenAI();

const myfile = await ai.files.upload({
  file: "path/to/sample.jpg",
  config: { mimeType: "image/jpeg" },
});

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: createUserContent([
    createPartFromUri(myfile.uri, myfile.mimeType),
    "Caption this image.",
  ]),
});
console.log(response.text);

IMAGE_PATH="cats-and-dogs.jpg"
MIME_TYPE=$(file -b --mime-type "${IMAGE_PATH}")
NUM_BYTES=$(wc -c < "${IMAGE_PATH}")
DISPLAY_NAME=IMAGE

tmp_header_file=upload-header.tmp

# Initial resumable request defining metadata.
curl "https://generativelanguage.googleapis.com/upload/v1beta/files" \
  -D upload-header.tmp \
  -H "X-Goog-Upload-Protocol: resumable" \
  -H "X-Goog-Upload-Command: start" \
  -H "X-Goog-Upload-Header-Content-Length: ${NUM_BYTES}" \
  -H "X-Goog-Upload-Header-Content-Type: ${MIME_TYPE}" \
  -H "Content-Type: application/json" \
  -d "{'file': {'display_name': '${DISPLAY_NAME}'}}" 2> /dev/null

upload_url=$(grep -i "x-goog-upload-url: " "${tmp_header_file}" | cut -d" " -f2 | tr -d "\r")
rm "${tmp_header_file}"

# Upload the actual bytes.
curl "${upload_url}" \
  -H "Content-Length: ${NUM_BYTES}" \
  -H "X-Goog-Upload-Offset: 0" \
  -H "X-Goog-Upload-Command: upload, finalize" \
  --data-binary "@${IMAGE_PATH}" 2> /dev/null > file_info.json

file_uri=$(jq -r ".file.uri" file_info.json)
echo file_uri=$file_uri

# Now generate content using that file
curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
    -H 'Content-Type: application/json' \
    -X POST \
    -d '{
      "contents": [{
        "parts":[
          {"file_data":{"mime_type": "'"${MIME_TYPE}"'", "file_uri": "'"${file_uri}"'"}},
          {"text": "Caption this image."}]
        }]
      }' 2> /dev/null > response.json

cat response.json
echo

jq -r ".candidates[].content.parts[].text" response.json

Learn more about Image Understanding:

Gemini API Image understanding (General introduction)
Gemma Image Understanding (Gemma-specific capabilities)

System Instructions¶

You can pass a system instruction to set the model's behavior:

PythonJavaScriptREST

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    config=types.GenerateContentConfig(
        system_instruction="You are a wise Kyoto tea master. Speak calmly and poetically, using nature metaphors. Keep answers under 3 sentences."
    ),
    contents="What is the purpose of the tea ceremony?"
)
print(response.text)

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

const ai = new GoogleGenAI();

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: "What is the purpose of the tea ceremony?",
  config: {
    systemInstruction: "You are a wise Kyoto tea master. Speak calmly and poetically, using nature metaphors. Keep answers under 3 sentences."
  }
});
console.log(response.text);

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [{
    "parts":[{"text": "What is the purpose of the tea ceremony?"}]
  }],
  "systemInstruction": {
    "parts": [{"text": "You are a wise Kyoto tea master. Speak calmly and poetically, using nature metaphors. Keep answers under 3 sentences."}]
  }
}'

Multi-turn Conversations¶

The SDK provides a chat interface that tracks conversation history automatically:

PythonJavaScriptREST

from google import genai

client = genai.Client()
chat = client.chats.create(model="gemma-4-26b-a4b-it")

response = chat.send_message("What are the three most famous castles in Japan?")
print(response.text)

response = chat.send_message("Which one should I visit in spring for cherry blossoms?")
print(response.text)

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

const ai = new GoogleGenAI();
const chat = ai.chats.create({ model: "gemma-4-26b-a4b-it" });

let response = await chat.sendMessage({ message: "What are the three most famous castles in Japan?" });
console.log(response.text);

response = await chat.sendMessage({ message: "Which one should I visit in spring for cherry blossoms?" });
console.log(response.text);

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [
    {
      "role": "user",
      "parts": [{ "text": "What are the three most famous castles in Japan?" }]
    },
    {
      "role": "model",
      "parts": [{ "text": "Himeji Castle, Matsumoto Castle, and Kumamoto Castle are often considered the top three." }]
    },
    {
      "role": "user",
      "parts": [{ "text": "Which one should I visit in spring for cherry blossoms?" }]
    }
  ]
}'

Function Calling¶

Define tools as function declarations. The model decides when to call them:

PythonJavaScriptREST

from google import genai
from google.genai import types

# Define the function declaration
get_weather = {
    "name": "get_weather",
    "description": "Get current weather for a given location.",
    "parameters": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "City and state, e.g. 'San Francisco, CA'",
            },
        },
        "required": ["location"],
    },
}

client = genai.Client()
tools = types.Tool(function_declarations=[get_weather])
config = types.GenerateContentConfig(tools=[tools])

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    contents="Should I bring an umbrella to Kyoto today?",
    config=config,
)

# The model returns a function call instead of text
if response.function_calls:
    for fc in response.function_calls:
        print(f"Function to call: {fc.name}")
        print(f"ID: {fc.id}")
        print(f"Arguments: {fc.args}")
else:
    print("No function call found in the response.")
    print(response.text)

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

const ai = new GoogleGenAI();

const get_weather = {
    name: "get_weather",
    description: "Get current weather for a given location.",
    parameters: {
        type: "object",
        properties: {
            location: {
                type: "string",
                description: "City and state, e.g. 'San Francisco, CA'",
            },
        },
        required: ["location"],
    },
};

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: "Should I bring an umbrella to Kyoto today?",
  config: {
    tools: [{ functionDeclarations: [get_weather] }]
  }
});

if (response.functionCalls) {
    for (const fc of response.functionCalls) {
        console.log(`Function to call: ${fc.name}`);
        console.log(`Arguments: ${JSON.stringify(fc.args)}`);
    }
} else {
    console.log("No function call found in the response.");
    console.log(response.text);
}

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [{
    "parts":[{"text": "Should I bring an umbrella to Kyoto today?"}]
  }],
  "tools": [{
    "functionDeclarations": [{
      "name": "get_weather",
      "description": "Get current weather for a given location.",
      "parameters": {
        "type": "OBJECT",
        "properties": {
          "location": {
            "type": "STRING",
            "description": "City and state, e.g. 'San Francisco, CA'"
          }
        },
        "required": ["location"]
      }
    }]
  }]
}'

Google Search¶

Ground Gemma 4 responses in real-time web data with Google Search:

PythonJavaScriptREST

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemma-4-26b-a4b-it",
    contents="What are the dates for cherry blossom season in Tokyo this year?",
    config=types.GenerateContentConfig(
        tools=[{"google_search":{}}]
    ),
)

print(response.text)

# Access grounding metadata for citations
for chunk in response.candidates[0].grounding_metadata.grounding_chunks:
    print(f"Source: {chunk.web.title} --- {chunk.web.uri}")

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

const ai = new GoogleGenAI();

const response = await ai.models.generateContent({
  model: "gemma-4-26b-a4b-it",
  contents: "What are the dates for cherry blossom season in Tokyo this year?",
  config: {
    tools: [{ googleSearch: {} }]
  }
});

console.log(response.text);

if (response.candidates?.[0]?.groundingMetadata?.groundingChunks) {
    for (const chunk of response.candidates[0].groundingMetadata.groundingChunks) {
        if (chunk.web) {
            console.log(`Source: ${chunk.web.title} --- ${chunk.web.uri}`);
        }
    }
}

curl "https://generativelanguage.googleapis.com/v1beta/models/gemma-4-26b-a4b-it:generateContent" \
-H 'Content-Type: application/json' \
-X POST \
-d '{
  "contents": [{
    "parts":[{"text": "What are the dates for cherry blossom season in Tokyo this year?"}]
  }],
  "tools": [{"googleSearch": {}}]
}'