Name: Google Gemma 3 4B IT Multimodal Chat LLM API
Brand: magicapi
SKU: api-magicapi-state-of-the-art-ai-google-deepmind-gemma-3-4b-it
Availability: InStock
Author: API Market

Shashank Agarwal

# 💬 Multimodal Chat LLM API (Google Gemma 3 4B IT)

## Overview
Gemma 3 4B IT is the 4B parameter, instruction-tuned member of Gemma 3, a family of open models from Google. Its developer describes Gemma as lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models.

Send a text `prompt` and, optionally, one `image` URL. Steer the model with `system_prompt` and control length and sampling with `max_new_tokens`, `temperature`, `top_p` and `top_k`. The reply comes back as a single string in `output`.

## Features
- **Text and image input**: ask about one `image` per request, or send text only.
- **System prompt**: `system_prompt` sets the assistant's role (default `You are a helpful assistant.`).
- **Output length**: `max_new_tokens` from 1 to 8192 (default `512`).
- **Sampling controls**: `temperature` 0 to 2 (default `0.7`), `top_p` 0 to 1 (default `0.9`), `top_k` 0 to 100 (default `50`).
- **Open model**: Gemma 3 is an open model family from Google.

## Use Cases
1. **Image description**: describe photos for alt text, catalogs or search indexes.
2. **Visual questions**: ask about what is in an image, such as objects, colors or text.
3. **Text tasks**: summarize, rewrite, classify or extract information from text in the prompt.
4. **Chat assistants**: answer user questions with a fixed `system_prompt` for tone and role.

## Visual Examples
### Example Input
![Example input: two giraffes in a grassy savanna with zebras in the background](https://blog.api.market/wp-content/uploads/2026/10/rmgb-input-api-market.jpg)

- `prompt`: "Describe this image in detail."
- `system_prompt`: "You are a helpful assistant."
- `max_new_tokens` `512`, `temperature` `0.7`, `top_p` `0.9`, `top_k` `50`

### Example Output
The reply, shortened (the full answer also covers the foreground, background, lighting and composition):

```text
Okay, here's a detailed description of the image you sent:

**Overall Impression:**

The image captures a serene and expansive scene in what appears to be the African savanna. It’s a medium shot, focusing on two giraffes and a glimpse of zebras in the background. The color palette is warm and muted, lending a slightly vintage or documentary feel.

...

Would you like me to focus on a specific aspect of the image, such as the giraffes’ anatomy, the landscape, or the lighting?<end_of_turn>
```

## Input Parameters
| Field | Type | Required | Default | Allowed values | Description |
| --- | --- | --- | --- | --- | --- |
| `prompt` | string | Yes | none |  | Text prompt for the model |
| `image` | string (URL) | No | none |  | Optional image URL for multimodal tasks (for example describing or asking about an image) |
| `system_prompt` | string | No | `You are a helpful assistant.` |  | System prompt to guide the model's behavior |
| `max_new_tokens` | integer | No | `512` | `1` to `8192` | Maximum number of tokens to generate |
| `temperature` | number | No | `0.7` | `0` to `2` | Sampling temperature |
| `top_p` | number | No | `0.9` | `0` to `1` | Top-p (nucleus) sampling |
| `top_k` | integer | No | `50` | `0` to `100` | Top-k sampling |

## Pricing
- **Pay As You Go (PAYG) pricing**: you only pay for what you use.
- **Price**: $0.0079 per API unit.
- **1 API unit = 1 request with up to 31 seconds of processing.** A request that needs more processing time is billed one more unit for each started 31-second block. Processing time grows with the length of the answer (`max_new_tokens`) and with image input.
- **Example cost**: the example request processed for 42.33 seconds, so it used 2 API units and cost $0.0158.
- **Failed or cancelled predictions are not billed.**
- **GET requests to check prediction status** are free and don't count toward API unit usage.
- **No monthly subscription fee.**

## API Documentation

### Base URL
All API requests go to: `https://prod.api.market/api/v1/magicapi/state-of-the-art-ai-google-deepmind-gemma-3-4b-it`

### Authentication
Send your api.market API key in the `x-api-market-key` header:
```
x-api-market-key: YOUR_API_MARKET_KEY
```

### Asynchronous Processing
**IMPORTANT**: This API is asynchronous. The flow is:
1. Send a POST request to `/predictions`. A successful request returns **201 Created** with a prediction `id` and a `status` of `starting` or `processing`.
2. Poll GET `/predictions/{id}` with that ID.
3. Keep polling while `status` is `starting` or `processing`.
4. Stop when `status` is `succeeded` (read `output`), `failed` or `canceled` (read `error`).

A polling interval of 1 second and an overall timeout of 600 seconds (10 minutes) work well.

### Endpoints

#### Create a Prediction
```bash
curl -X POST "https://prod.api.market/api/v1/magicapi/state-of-the-art-ai-google-deepmind-gemma-3-4b-it/predictions" \
  -H "Content-Type: application/json" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY" \
  -d '{
    "version": "00139d2960396352b671f7b5c2ece5313bf6d45fe0a052efe14f023d2a81e196",
    "input": {
      "image": "https://blog.api.market/wp-content/uploads/2026/10/rmgb-input-api-market.jpg",
      "prompt": "Describe this image in detail.",
      "system_prompt": "You are a helpful assistant.",
      "max_new_tokens": 512,
      "temperature": 0.7,
      "top_p": 0.9,
      "top_k": 50
    }
  }'
```

Response (201 Created):
```json
{
  "id": "yd03jytx1drge0d1mvjr9enz2w",
  "version": "00139d2960396352b671f7b5c2ece5313bf6d45fe0a052efe14f023d2a81e196",
  "status": "starting",
  "created_at": "2026-11-04T14:10:11.851Z"
}
```

#### Check Prediction Status
```bash
curl -X GET "https://prod.api.market/api/v1/magicapi/state-of-the-art-ai-google-deepmind-gemma-3-4b-it/predictions/{id}" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

Response when the prediction has finished (200 OK):
```json
{
  "id": "yd03jytx1drge0d1mvjr9enz2w",
  "status": "succeeded",
  "output": "Okay, here's a detailed description of the image you sent:\n\n**Overall Impression:**\n\nThe image captures a serene and expansive scene in what appears to be the African savanna. It’s a medium shot, focusing on two giraffes and a glimpse of zebras in the background. The color palette is warm and muted, lending a slightly vintage or documentary feel.\n\n...\n\nWould you like me to focus on a specific aspect of the image, such as the giraffes’ anatomy, the landscape, or the lighting?<end_of_turn>",
  "error": null,
  "created_at": "2026-11-04T14:10:11.851Z",
  "started_at": "2026-11-04T14:13:22.338677274Z",
  "completed_at": "2026-11-04T14:14:04.667025205Z"
}
```

`output` is a single string with the model's whole reply (shortened above). The reply can end with the literal marker `<end_of_turn>`; remove it before showing the text.

### Status Codes
- **201 Created**: prediction created (POST `/predictions`). This is the normal success response.
- **200 OK**: prediction details returned (GET `/predictions/{id}`).
- **400 Bad Request**: invalid input, for example a missing `prompt`, a `max_new_tokens` above 8192 or a `temperature` above 2.
- **401 Unauthorized**: missing or invalid API key.
- **404 Not Found**: no prediction with that ID.
- **429 Too Many Requests**: rate limit exceeded.
- **500 Internal Server Error** or **503 Service Unavailable**: temporary server problem; retry later.

## Rate Limits
Requests are rate limited to keep usage fair. If you receive a 429 response, wait and retry with exponential backoff.

## Error Handling
- Check the HTTP status code of every request and read the response body for the error message.
- A prediction can also end with `status` set to `failed`; the `error` field then explains why. Failed predictions are not billed.
- Retry 429, 500 and 503 responses with exponential backoff. Fix the input before retrying a 400.

## Implementation Guide
### Example Code

#### JavaScript
```javascript
const apiKey = "YOUR_API_MARKET_KEY";
const baseUrl = "https://prod.api.market/api/v1/magicapi/state-of-the-art-ai-google-deepmind-gemma-3-4b-it";

async function createAndWaitForPrediction(input) {
  const response = await fetch(`${baseUrl}/predictions`, {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-api-market-key": apiKey
    },
    body: JSON.stringify({ version: "00139d2960396352b671f7b5c2ece5313bf6d45fe0a052efe14f023d2a81e196", input })
  });

  if (response.status !== 201) {
    throw new Error(`API error: ${response.status} ${await response.text()}`);
  }

  const prediction = await response.json();
  return pollForResults(prediction.id);
}

const delay = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function pollForResults(predictionId) {
  const maxTimeout = 600000;
  const pollInterval = 1000;
  const startTime = Date.now();

  while (Date.now() - startTime < maxTimeout) {
    const response = await fetch(`${baseUrl}/predictions/${predictionId}`, {
      headers: { "x-api-market-key": apiKey }
    });

    if (!response.ok) {
      throw new Error(`API error: ${response.status} ${await response.text()}`);
    }

    const prediction = await response.json();

    if (prediction.status === "succeeded") {
      return prediction.output;
    }

    if (prediction.status === "failed" || prediction.status === "canceled") {
      throw new Error(`Prediction ${prediction.status}: ${prediction.error || "Unknown error"}`);
    }

    await delay(pollInterval);
  }

  throw new Error("Prediction timed out after 10 minutes");
}

createAndWaitForPrediction({
  "image": "https://blog.api.market/wp-content/uploads/2026/10/rmgb-input-api-market.jpg",
  "prompt": "Describe this image in detail.",
  "system_prompt": "You are a helpful assistant.",
  "max_new_tokens": 512,
  "temperature": 0.7,
  "top_p": 0.9,
  "top_k": 50
}).then((output) => console.log(output.replace("<end_of_turn>", "").trim()));
```

#### Python
```python
import time
import requests

API_KEY = "YOUR_API_MARKET_KEY"
BASE_URL = "https://prod.api.market/api/v1/magicapi/state-of-the-art-ai-google-deepmind-gemma-3-4b-it"


def create_and_wait_for_prediction(input_data):
    headers = {"Content-Type": "application/json", "x-api-market-key": API_KEY}
    response = requests.post(
        f"{BASE_URL}/predictions",
        headers=headers,
        json={"version": "00139d2960396352b671f7b5c2ece5313bf6d45fe0a052efe14f023d2a81e196", "input": input_data},
    )

    if response.status_code != 201:
        raise Exception(f"API error: {response.status_code} {response.text}")

    return poll_for_results(response.json()["id"])


def poll_for_results(prediction_id):
    max_timeout = 600
    poll_interval = 1
    headers = {"x-api-market-key": API_KEY}
    start_time = time.time()

    while time.time() - start_time < max_timeout:
        response = requests.get(f"{BASE_URL}/predictions/{prediction_id}", headers=headers)

        if response.status_code != 200:
            raise Exception(f"API error: {response.status_code} {response.text}")

        prediction = response.json()

        if prediction["status"] == "succeeded":
            return prediction["output"]

        if prediction["status"] in ("failed", "canceled"):
            raise Exception(f"Prediction {prediction['status']}: {prediction.get('error') or 'Unknown error'}")

        time.sleep(poll_interval)

    raise Exception("Prediction timed out after 10 minutes")


print(create_and_wait_for_prediction({
    "image": "https://blog.api.market/wp-content/uploads/2026/10/rmgb-input-api-market.jpg",
    "prompt": "Describe this image in detail.",
    "system_prompt": "You are a helpful assistant.",
    "max_new_tokens": 512,
    "temperature": 0.7,
    "top_p": 0.9,
    "top_k": 50
}).replace("<end_of_turn>", "").strip())
```

## FAQ
- **Can I send an image?**
  - Yes. Put one image URL in `image` and ask about it in `prompt`. Leave `image` out for text-only requests.
- **How long can the answer be?**
  - Up to `max_new_tokens` tokens, from 1 to 8192 (default 512). Longer answers take longer to generate and can use more API units.
- **Why does the answer end with `<end_of_turn>`?**
  - That is the model's end-of-reply marker. It is part of the returned string; strip it before display, as the code examples do.
- **How do I get more predictable answers?**
  - Lower `temperature` (for example 0.2) and keep `system_prompt` specific. Higher values give more varied text.
- **How is a request billed?**
  - By processing time: $0.0079 per started block of 31 seconds. The example processed for 42.33 seconds and cost $0.0158.
- **What if my prediction fails?**
  - Failed predictions are not billed. Read the `error` field, adjust the input and try again.

## Contact Us
If you have any questions, reach out to us through the chat on the bottom right of the page.


Run Google's open Gemma 3 4B IT model: answer prompts, describe images and follow a system prompt. $0.0079 per request with up to 31 seconds of processing, pay as you go.