[Models](/models)/[Captioning](/models?category=captioning)/google/gemma-4-12B-it

HFScene CaptioningApache 2.0

# gemma-4-12B-it

by google

Open 12B multimodal model for image, audio, and long-context agent perception

581Kdl/month

12Bparams

[HuggingFace](https://huggingface.co/google/gemma-4-12B-it)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`google/gemma-4-12B-it`

## Deploy gemma-4-12B-it

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

Gemma 4 12B IT is an instruction-tuned open model from Google DeepMind. The model card describes Gemma 4 as multimodal, with text and image input across the family and audio support on the E2B, E4B, and 12B variants. It is a strong fit for agents that need to inspect retrieved images, short audio clips, or mixed evidence after first-stage search.

On Mixpeek, Gemma 4 12B belongs in the inspection layer. Use cheaper embeddings and filters to retrieve candidates, then ask Gemma to produce concise observations, answer bounded visual questions, or turn multimodal evidence into structured fields that downstream agents can cite.

## Architecture

Instruction-tuned Gemma 4 multimodal model exposed through Hugging Face Transformers. The 12B checkpoint supports a 256K context window, multilingual text handling, image input, audio input, and text output.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so gemma-4-12B-it runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace\_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "asset-00412",
          // The model produces text, so it lands in payload. Give the
          // collection a text vector index and embed that text to make it
          // searchable rather than only filterable.
          payload: { extracted\_text: modelOutput, source\_key: "archive/2026/asset-00412" },
          vectors: { "multimodal-embedding": embeddingOfModelOutput },
        },
      \],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// universal\_extractor@v1 runs google/gemini-embedding-2
// (3072-d) over a bucket, with no inference of your own.

## Capabilities

*   Image-text and audio-text understanding in a single instruction-tuned model
*   Long-context multimodal reasoning for evidence inspection
*   Multilingual support across broad language coverage
*   Apache 2.0 license for production evaluation

## Use Cases on Mixpeek

Answer visual questions over retrieved product, ad, or scene images

Summarize mixed evidence that includes transcript snippets and frames

Extract structured observations from screenshots, charts, and media frames

Run second-stage agent inspection after vector search narrows the candidate set

## Performance

Input SizeText, image, and audio inputs with long-context text

GPU LatencyOutput length dependent

GPU ThroughputBatch dependent

GPU Memory12B multimodal deployment class

Best used as a second-stage inspector after retrieval narrows candidates

## Common Pipeline Companions

[

openai/clip-vit-large-patch14

Fast first-stage visual retrieval

](/model/openai/clip-vit-large-patch14)[

Qwen/Qwen3-VL-Reranker-2B

Rerank visual evidence before inspection

](/model/Qwen/Qwen3-VL-Reranker-2B)[

nvidia/nemotron-3.5-asr-streaming-0.6b

Generate transcript evidence for audio-aware inspection

](/model/nvidia/nemotron-3.5-asr-streaming-0.6b)

## Explore on Mixpeek

[

More Captioning models

Compare alternatives in this category

](/models?category=captioning)[

Best Document AI Platforms

Hand-picked tools & platforms compared

](/curated-lists/best-document-ai-platforms)[

OCR & Document AI Internals

Deep-dive technical guide

](/guides/ocr-document-ai-internals)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

FrameworkHF

Organizationgoogle

FeatureScene Captioning

Outputtext

Modalitiesvideo, image

RetrieverSemantic Search

Parameters12B

LicenseApache 2.0

Downloads/mo581K

[

### Research Paper

Gemma 4 12B IT model card

arxiv.org](https://huggingface.co/google/gemma-4-12B-it)

### Build a pipeline with gemma-4-12B-it

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

### Alternative Models

[

Salesforce/blip2-opt-2.7b

Scene Captioning

](/model/Salesforce/blip2-opt-2.7b)[

microsoft/Florence-2-large

Scene Captioning

](/model/microsoft/Florence-2-large)[

google/paligemma2-3b-mix-448

Scene Captioning

](/model/google/paligemma2-3b-mix-448)[

Qwen/Qwen3-VL-8B-Instruct

Scene Captioning

](/model/Qwen/Qwen3-VL-8B-Instruct)