# Gemma 4 26B-A4B

> Canonical: https://www.overmindlab.ai/models/gemma-4-26b-a4b-it

Best speed/quality MoE — 26B total, ~4B active, 256K context.

| Property | Value |
| --- | --- |
| Model ID | google/gemma-4-26B-A4B-it |
| Provider | Google |
| Group | Gemma 4 |
| Parameters | 26B (4B active) |
| Tier | Large |
| Context window | 256K |
| Max training context | 256K |
| Tool calling | Yes |
| Training methods | — |
| Training cost | from $2.00 |
| Serving cost (per 1M output tokens) | from $8.00 |

## Overview

Gemma 4 26B-A4B is the MoE instruct checkpoint for computer-use style jobs: strong quality with active-param serve cost closer to a mid model. Prefer over 31B when latency and VRAM matter more than peak scores.

Google DeepMind Gemma 4 open models: hybrid thinking, 140+ languages, dense and MoE sizes from edge to server.

## Good for

- Agentic computer-use and long-context coding
- QLoRA when dense 31B is too heavy
- Multimodal text+image production agents

Train this model on your agent's data: https://docs.overmindlab.ai/models/training.md

## Related models

- [Gemma 4 E4B](https://www.overmindlab.ai/models/gemma-4-e4b-it): E4B (~8B), 128K context
- [Gemma 4 E2B](https://www.overmindlab.ai/models/gemma-4-e2b-it): E2B (~5.1B), 128K context
- [Gemma 4 12B](https://www.overmindlab.ai/models/gemma-4-12b-it): 12B, 256K context
- [Gemma 4 31B](https://www.overmindlab.ai/models/gemma-4-31b-it): 31B, 256K context