[Models](/models)/[Captioning](/models?category=captioning)/google/gemma-4-26B-A4B-it

HFScene CaptioningApache-2.0

# gemma-4-26B-A4B-it

by google

Mixture-of-experts VLM delivering 97% of 31B quality at 8x less compute

1.1Mdl/month

26B total / 4B activeparams

[HuggingFace](https://huggingface.co/google/gemma-4-26B-A4B-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-4-26B-A4B-it`

## Deploy gemma-4-26B-A4B-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 27B-A4B is Google's MoE vision-language model that activates only 4B parameters per token from a total of 26B. It ranked #6 on the Arena AI leaderboard at launch while using a fraction of the compute of dense models its size.

The model handles both text and image input with a 256K context window, making it suitable for long-document visual understanding. Its efficiency profile makes it the best choice when you need high-quality VLM capabilities at manageable cost.

## Architecture

Mixture-of-Experts architecture with 26B total parameters, 4B active per token. Vision encoder processes image patches alongside text tokens. 256K context window. Supports optional 'thinking' mode for chain-of-thought reasoning.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so gemma-4-26B-A4B-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

*   Multimodal understanding (text + images)
*   256K context window for long documents
*   MoE efficiency: 4B active / 26B total
*   Built-in reasoning mode
*   Apache 2.0 license

## Use Cases on Mixpeek

Cost-efficient visual document captioning in Mixpeek ingestion pipelines

Long-document visual understanding (multi-page PDFs with charts)

Scene description for video frame analysis at scale

## Benchmarks

Dataset

Metric

Score

Source

MMLU Pro

Accuracy

83%

Google, May 2026

AIME 2026

Accuracy

85%

Google, May 2026

Arena AI Leaderboard

ELO

1441 (#6)

Arena AI, May 2026

## Performance

Input SizeUp to 256K tokens (text + image patches)

GPU Latency~120ms / image (A100, 4B active)

GPU Throughput~65 images/sec (A100, batch 8)

GPU Memory~18 GB (MoE, sparse activation)

## Common Pipeline Companions

[

BAAI/bge-m3

Embed captions for semantic search

](/model/BAAI/bge-m3)[

cross-encoder/ettin-reranker-1b-v1

Rerank caption search results

](/model/cross-encoder/ettin-reranker-1b-v1)

## 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

Parameters26B total / 4B active

LicenseApache-2.0

Downloads/mo1.1M

[

### Research Paper

Gemma 4: Byte for byte, the most capable open models

arxiv.org](https://deepmind.google/models/gemma/gemma-4/)

### Build a pipeline with gemma-4-26B-A4B-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)