Name: Gemma 4 26B A4B Fast API: LLM with Image and Video
Brand: magicapi
SKU: api-magicapi-multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast
Availability: InStock
Author: API Market

Shashank Agarwal

# 💬 LLM Chat API with Image and Video Input (Google Gemma 4 26B A4B Fast)

## Overview
This API runs Google's Gemma 4 26B A4B, a mixture-of-experts (MoE) language model, in a version optimized for speed by Pruna AI. Send a text `message`, optionally with `images` or a `video`, and the model replies with text.

The reply is streamed as a list of text chunks; join them to get the full answer. An optional thinking mode lets the model reason internally before it answers.

## Features
- **Text chat**: send the user turn in `message` and set behaviour with `system_prompt`.
- **Image input**: pass one or more image URLs in `images`. `max_visual_tokens` sets the token budget per image (70, 140, 280, 560 or 1120; higher means more detail and more compute).
- **Video input**: pass a `video` URL; `video_fps` (0.1 to 30, default 2) sets how many frames per second are sampled.
- **Thinking mode**: set `enable_thinking` to `true` to let the model reason internally before answering.
- **Long replies**: `max_tokens` from 1 to 16384 (default 2048).
- **Sampling controls**: `temperature` (0 to 2), `top_p` (0 to 1) and `presence_penalty` (1 to 1.5).

## Use Cases
1. **Chat assistants and Q&A**: answer user questions in apps, help desks and internal tools.
2. **Image questions**: describe photos, read screenshots or compare several images in one request.
3. **Video questions**: summarize or answer questions about a short video.
4. **Text tasks**: summarize, rewrite, classify or extract information from text.

## Visual Examples
### Example Input
Message: "Explain vLLM in one sentence"

Settings: no images or video, `max_tokens` `2048`, `temperature` `0.6`, `top_p` `0.95`, `presence_penalty` `1`, `enable_thinking` `false`, `video_fps` `2`, `max_visual_tokens` `280`.

### Example Output
The output is a list of 35 text chunks. Joined, they read:

> vLLM is a high-throughput, memory-efficient serving engine for large language models that utilizes **PagedAttention** to optimize KV cache management and significantly accelerate inference speeds.

## Input Parameters
| Field | Type | Required | Default | Allowed values | Description |
| --- | --- | --- | --- | --- | --- |
| `message` | string | No | `Explain vLLM in one sentence` |  | The user message to send to the model |
| `images` | array of strings (URLs) | No | `[]` |  | Images to send to the model |
| `video` | string (URL) | No | none |  | Video file to send to the model |
| `video_fps` | number | No | `2` | 0.1 to 30 | Frames per second to sample from the video |
| `system_prompt` | string | No | none |  | Optional system prompt to set the model's behavior |
| `max_tokens` | integer | No | `2048` | 1 to 16384 | Maximum number of tokens to generate |
| `enable_thinking` | boolean | No | `false` |  | Enable thinking mode (the model reasons internally before answering) |
| `temperature` | number | No | `0.6` | 0 to 2 | Sampling temperature (higher is more creative, lower is more deterministic) |
| `top_p` | number | No | `0.95` | 0 to 1 | Nucleus sampling: only consider tokens with cumulative probability up to this value |
| `presence_penalty` | number | No | `1` | 1 to 1.5 | Presence penalty: penalize new tokens based on their presence in the text |
| `max_visual_tokens` | integer | No | `280` | 70 to 1120; supported values 70, 140, 280, 560, 1120 | Vision token budget per image (higher means more detail and more compute) |

## Pricing
- **Pay As You Go (PAYG) pricing**: you only pay for what you use.
- **Price**: $0.0068 per API unit.
- **1 API unit = 1 request with up to 4 seconds of processing.** A request that takes longer is billed one more unit for each started 4-second block (for example, 10 seconds of processing is 3 units, $0.0204).
- Billing is by processing time, not by tokens. Long replies (`max_tokens`), `enable_thinking`, many images, high `max_visual_tokens` and videos take longer and can add units.
- The example request (a one-sentence answer) took 1.9 seconds of processing, so it was billed 1 unit ($0.0068).
- **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/multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast`

### 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/multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast/predictions" \
  -H "Content-Type: application/json" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY" \
  -d '{
    "version": "007dac3717afbfb1ddade995c5fbff003a5a365660b8f8155b6bbb053eada6e4",
    "input": {
      "top_p": 0.95,
      "images": [],
      "message": "Explain vLLM in one sentence",
      "video_fps": 2,
      "max_tokens": 2048,
      "temperature": 0.6,
      "enable_thinking": false,
      "presence_penalty": 1,
      "max_visual_tokens": 280
    }
  }'
```

Response (201 Created):
```json
{
  "id": "5mfac14frhrmw0d1d8n9cmp1m4",
  "version": "007dac3717afbfb1ddade995c5fbff003a5a365660b8f8155b6bbb053eada6e4",
  "status": "starting",
  "created_at": "2026-10-23T19:08:56.798Z"
}
```

#### Check Prediction Status
```bash
curl -X GET "https://prod.api.market/api/v1/magicapi/multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast/predictions/{id}" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

Response when the prediction has finished (200 OK):
```json
{
  "id": "5mfac14frhrmw0d1d8n9cmp1m4",
  "status": "succeeded",
  "output": ["v", "LL", "M", " is", " a", " high", "-", "throughput", ",", " memory", "-", "efficient", " serving", " engine", " for", " large", " language", " models", " that", " utilizes", " **", "Paged", "Attention", "**", " to", " optimize", " KV", " cache", " management", " and", " significantly", " accelerate", " inference", " speeds", "."],
  "error": null,
  "created_at": "2026-10-23T19:08:56.798Z",
  "started_at": "2026-10-23T19:08:57.798Z",
  "completed_at": "2026-10-23T19:08:59.798Z"
}
```

`output` is an array of text chunks. Join them in order with no separator (`output.join("")`) to get the reply.

### 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 `max_tokens` above 16384 or `presence_penalty` below 1.
- **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
All plans allow 10 requests per second. Requests above that return `429 Too Many Requests`; back off before retrying and honour any `Retry-After` header. Polling a prediction with GET is free but still counts toward the per-second limit, so poll every 1 to 5 seconds rather than in a tight loop.

## 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 (for example, an image or video URL that cannot be downloaded). 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/multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast";

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: "007dac3717afbfb1ddade995c5fbff003a5a365660b8f8155b6bbb053eada6e4", 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({
  top_p: 0.95,
  images: [],
  message: "Explain vLLM in one sentence",
  video_fps: 2,
  max_tokens: 2048,
  temperature: 0.6,
  enable_thinking: false,
  presence_penalty: 1,
  max_visual_tokens: 280
}).then((output) => console.log(output.join("")));
```

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

API_KEY = "YOUR_API_MARKET_KEY"
BASE_URL = "https://prod.api.market/api/v1/magicapi/multi-task-ai-inference-prunaai-gemma-4-26b-a4b-fast"


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": "007dac3717afbfb1ddade995c5fbff003a5a365660b8f8155b6bbb053eada6e4", "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")


output = create_and_wait_for_prediction({
    "top_p": 0.95,
    "images": [],
    "message": "Explain vLLM in one sentence",
    "video_fps": 2,
    "max_tokens": 2048,
    "temperature": 0.6,
    "enable_thinking": False,
    "presence_penalty": 1,
    "max_visual_tokens": 280,
})
print("".join(output))
```

## FAQ
- **Why is the output a list?**
  - The reply is returned as text chunks in order. Join them with no separator to get the full text.
- **How do I send images?**
  - Put one or more image URLs in `images`. Raise `max_visual_tokens` (up to 1120) when the model needs to read fine detail such as small text; lower it to save processing time.
- **How do I send a video?**
  - Pass the URL in `video`. `video_fps` controls how many frames per second the model looks at (default 2).
- **What does `enable_thinking` do?**
  - It lets the model reason internally before it answers. This can help on harder questions but takes longer, so it can use more units.
- **How long can replies be?**
  - Up to 16384 tokens, set with `max_tokens` (default 2048).
- **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. You can also email our support team at hello@magicapi.com.


Run Google's Gemma 4 26B A4B MoE model in Pruna AI's fast version: text chat with image and video input, thinking mode and up to 16384 output tokens.