Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.gmicloud.ai/llms.txt

Use this file to discover all available pages before exploring further.

Model ID 
gemini-batch

​
Gemini Batch Inference API Usage Guide

​
Overview

Gemini Batch Inference allows you to process large volumes of requests asynchronously at approximately 50% lower cost than online inference. Ideal for batch processing tasks like document analysis, image labeling, or bulk content generation.

​
Authentication

All API requests require authentication using an API key. Include your API key in the Authorization header: 
Authorization: Bearer YOUR_API_KEY

​
Submit Batch Job

​
Base URL

https://console.gmicloud.ai

​
Endpoint

POST /api/v1/ie/requestqueue/apikey/requests

​
Request Format

curl -X POST "https://console.gmicloud.ai/api/v1/ie/requestqueue/apikey/requests" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-batch",
    "payload": {
      "model": "gemini-3-flash-preview",
      "input_data": "{\"request\":{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"What is 2+2?\"}]}]}}\n{\"request\":{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"What is the capital of France?\"}]}]}}"
    }
  }'

​
Request Parameters

ParameterTypeRequiredDescriptionDefaultConstraints
modelenumYesTarget Gemini model for batch prediction.gemini-3-flash-previewSee supported models below
input_datastringYesJSONL content where each line is a request object. Can be raw JSONL string or base64 encoded.-Max 1GB file size, max 200K requests

​
Supported Models

ModelDescriptionBest For
gemini-3-flash-previewFast, efficient Gemini 3 FlashHigh-volume, simple tasks
gemini-3-pro-previewMost capable Gemini 3 ProComplex reasoning tasks
gemini-3-pro-image-previewGemini 3 Pro with image generationBatch image generation

​
JSONL Input Format

Each line must be a valid JSON object with a request field containing contents: Text-only requests: 
{"request":{"contents":[{"role":"user","parts":[{"text":"What is the capital of France?"}]}]}}
{"request":{"contents":[{"role":"user","parts":[{"text":"Summarize quantum computing in 3 sentences."}]}]}}
{"request":{"contents":[{"role":"user","parts":[{"text":"Write a haiku about mountains."}]}]}}
Multimodal requests (with images): 
{"request":{"contents":[{"role":"user","parts":[{"text":"Describe this image"},{"file_data":{"file_uri":"gs://bucket/image.jpg","mime_type":"image/jpeg"}}]}]}}

​
Response (Immediate)

After submitting, you receive a request_id to track the job: 
{
  "request_id": "7eaa77fc-bc67-4021-9f1b-96b3fd832314",
  "model": "gemini-batch",
  "status": "processing",
  "outcome": {
    "batch_job_state": "JOB_STATE_QUEUED"
  },
  "created_at": 1761763441,
  "updated_at": 1761763441
}

​
Check Job Status

​
Endpoint

GET /api/v1/ie/requestqueue/apikey/requests/{request_id}

​
Example

curl -X GET "https://console.gmicloud.ai/api/v1/ie/requestqueue/apikey/requests/7eaa77fc-bc67-4021-9f1b-96b3fd832314" \
  -H "Authorization: Bearer YOUR_API_KEY"

​
Response (In Progress)

{
  "request_id": "7eaa77fc-bc67-4021-9f1b-96b3fd832314",
  "model": "gemini-batch",
  "status": "processing",
  "outcome": {
    "batch_job_state": "JOB_STATE_RUNNING"
  },
  "created_at": 1761763441,
  "updated_at": 1761764441
}

​
Response (Completed)

{
  "request_id": "7eaa77fc-bc67-4021-9f1b-96b3fd832314",
  "model": "gemini-batch",
  "status": "success",
  "outcome": {
    "batch_job_state": "JOB_STATE_SUCCEEDED",
    "output_url": "https://storage.googleapis.com/.../predictions.jsonl",
    "output_download_urls": [
      "https://storage.googleapis.com/.../predictions-00000-of-00001.jsonl"
    ],
    "batch_job_completion_stats": {
      "successful_count": "100",
      "failed_count": "2"
    },
    "token_usage": {
      "total_prompt_tokens": 5000,
      "total_candidates_tokens": 8000,
      "successful_requests": 100,
      "failed_requests": 2
    },
    "actual_cost_usd": "$0.001234"
  },
  "created_at": 1761763441,
  "updated_at": 1761765441
}

​
Download and Parse Output

The output is a JSONL file where each line corresponds to one input request: 
{
  "status": "",
  "processed_time": "2024-01-15T10:30:00.000+00:00",
  "request": {"contents": [{"role": "user", "parts": [{"text": "What is 2+2?"}]}]},
  "response": {
    "candidates": [{
      "content": {"parts": [{"text": "4"}], "role": "model"},
      "finishReason": "STOP"
    }],
    "usageMetadata": {
      "promptTokenCount": 5,
      "candidatesTokenCount": 1,
      "totalTokenCount": 6
    }
  }
}
Note: An empty status field indicates success. Failed requests will have an error message in status.

​
List Your Batch Jobs

​
Endpoint

GET /api/v1/ie/requestqueue/apikey/requests?model_id=gemini-batch

​
Example

curl -X GET "https://console.gmicloud.ai/api/v1/ie/requestqueue/apikey/requests?model_id=gemini-batch" \
  -H "Authorization: Bearer YOUR_API_KEY"

​
Request Status Values

StatusDescription
queuedJob is waiting to be submitted to Vertex AI
processingBatch job is running (may take minutes to hours)
successJob completed (check batch_job_completion_stats for details)
failedJob failed (check error message in outcome)

​
Batch Job States (Vertex AI)

StateDescription
JOB_STATE_QUEUEDWaiting for resources
JOB_STATE_PENDINGJob is being prepared
JOB_STATE_RUNNINGProcessing requests
JOB_STATE_SUCCEEDEDAll requests completed
JOB_STATE_PARTIALLY_SUCCEEDEDSome requests failed
JOB_STATE_FAILEDJob failed
JOB_STATE_CANCELLEDJob was cancelled (e.g., 24h timeout)

​
Limits

LimitValue
Max input file size1 GB
Max requests per job200,000
Max processing time24 hours (after job starts running)
Max queue time72 hours

​
Tips for Best Results

  1. Batch Size: Ideal for 100+ requests. For fewer requests, consider online inference.
  2. File References: Use gs:// URIs for images/documents stored in Google Cloud Storage.
  3. Processing Time: Jobs typically complete within minutes to hours depending on volume.
  4. Cost Optimization: Use batch for non-time-sensitive workloads to save ~50%.
  5. Error Handling: Check batch_job_completion_stats for failed request counts.