As of: 2026-11-13 13:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/gemma-4/

# Gemma 4

Google DeepMind, Gemma family. Status: current. Type: language model. Released: 2026-04-02. Training cutoff: not published.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face | `google/gemma-4-31B-it` |  | https://huggingface.co/google/gemma-4-31B-it |
| Hugging Face | `google/gemma-4-26B-A4B-it` |  | https://huggingface.co/google/gemma-4-26B-A4B-it |
| Hugging Face | `google/gemma-4-12B-it` |  | https://huggingface.co/google/gemma-4-12B-it |
| Hugging Face | `google/gemma-4-E4B-it` |  | https://huggingface.co/google/gemma-4-E4B-it |
| Hugging Face | `google/gemma-4-E2B-it` |  | https://huggingface.co/google/gemma-4-E2B-it |
| OpenRouter | `google/gemma-4-31b-it` |  |  |
| OpenRouter | `google/gemma-4-26b-a4b-it` |  |  |
| Google docs |  |  | https://ai.google.dev/gemma/docs/core |

## Pricing

- Input: $0.09 per 1M tokens
- Output: $0.34 per 1M tokens
- Unit as published: per 1M tokens, OpenRouter price for google/gemma-4-31b-it (26B-A4B: $0.09 / $0.30; free variants exist). Weights free to download
- Source: https://openrouter.ai/google/gemma-4-31b-it
- Last checked: 2026-09-29

## Specs

- Input: text, image, audio, video
- Output: text
- Context window: 262,144 tokens
- Max output: not published
- Open weights: yes
- Licence: apache-2.0

## What it has not seen

- No published training cutoff. Counted from its release (2026-04-02): at least 1,887 AI events since (432 major, 85 historic), as of 2026-11-13.
- Everything since March 2026: https://postcutoff.com/since/2026-03/

## What stands out

- First Apache-2.0 Gemma: First Gemma generation under the permissive Apache 2.0 license instead of Google's custom Gemma terms. Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
- Intelligence per parameter: 31B dense ranked #3 and 26B A4B MoE #6 among open models on Arena at launch. Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
- On-device agentic models: E2B/E4B edge models with native audio+vision, function calling and structured JSON, running offline on phones/Raspberry Pi/Jetson. Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
- Encoder-free unified 12B (found after launch): Gemma 4 12B, added later, is a unified encoder-free multimodal model with native audio. Source: https://ai.google.dev/gemma/docs/core

## Notes

Sizes E2B, E4B (128K context), 12B, 26B A4B MoE, 31B dense (256K context); base and -it variants plus QAT/GGUF quantized repos. 12B released later (HF repo 2026-05-23). Audio input on E2B/E4B/12B only. 140+ languages. pricing is third-party (OpenRouter), not Google.

Latest open-weights family from Google DeepMind (built from Gemini 3 research); run locally or self-host.

```python
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="google/gemma-4-E4B-it")
print(pipe(text=[{"role":"user","content":[{"type":"text","text":"Hello!"}]}]))
```

Sources: [Gemma docs](https://ai.google.dev/gemma/docs/core), [launch blog](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/), [HF repo](https://huggingface.co/google/gemma-4-31B-it).

Model page: https://postcutoff.com/m/gemma-4/