# Gemma 4

> **TL;DR.** Gemma 4 is Google DeepMind's fourth-generation family of open-weight multimodal models, released April 2, 2026 under Apache 2.

- **Category:** AI / Open Models / Multimodal
- **Stage:** validating
- **Age:** 62 days
- **Origin date:** 2026-04-02
- **First detected:** 2026-04-28
- **Canonical URL:** https://earlyterms.com/term/gemma-4
- **Sources:** 7 primary URLs

## Definition

Gemma 4 is Google DeepMind's fourth-generation family of open-weight multimodal models, released April 2, 2026 under Apache 2.0. Four sizes span phones to data centers: E2B, E4B, 26B Mixture-of-Experts, and a 31B dense model — all natively processing text, images, and video.

The [April 2 launch](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) positioned Gemma 4 as the most capable commercially-permissive open model family. The 31B model ranked #3 on the LMArena open-model leaderboard, AIME 2026 math scoring jumped from 20.8% (Gemma 3) to 89.2%, and the Gemmaverse community has now generated over 100,000 model variants.

## Analogy

Think of it as Android for AI models: free to deploy anywhere, optimized for every screen size.

## Why it's emerging now

Google released Gemma 4 on April 2, 2026 with four models (E2B through 31B) under Apache 2.0, making frontier-grade multimodal inference available entirely offline — on iPhones, MacBooks, and edge servers. Multi-token prediction drafters shipped May 5, delivering 3x inference speedups without quality loss, extending the model family's lifecycle well past its launch.

## Related terms

- *competitor:* Qwen 3.6
- *competitor:* qwen3
- *competitor:* qwen3-6
- *related:* Gemini 3.1 Pro
- *related:* gemini-3-1-pro
- *related:* MLX
- *related:* mlx
- *related:* LM Studio
- *related:* lm-studio
- *child:* MTP
- *child:* mtp
- *parent:* agentic-coding
- *competitor:* Llama 3
- *competitor:* Mistral

## Sources

1. [Google Blog — Gemma 4: Byte for byte, the most capable open models (Apr 2, 2026)](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/)
2. [Google AI for Developers — Gemma 4 model overview (architecture, specs, context windows)](https://ai.google.dev/gemma/docs/core)
3. [Hugging Face — Welcome Gemma 4: Frontier multimodal intelligence on device](https://huggingface.co/blog/gemma4)
4. [Google Blog — Accelerating Gemma 4: faster inference with multi-token prediction drafters (May 5, 2026)](https://blog.google/innovation-and-ai/technology/developers-tools/multi-token-prediction-gemma-4/)
5. [HN — Google releases Gemma 4 open models (1,812 pts, Apr 2, 2026)](https://news.ycombinator.com/item?id=47616361)
6. [HN — Indexing a year of video locally on a 2021 MacBook with Gemma4-31B (470 pts, May 21, 2026)](https://news.ycombinator.com/item?id=48222733)
7. [Interconnects.ai — Gemma 4 and what makes an open model succeed (Nathan Lambert analysis)](https://www.interconnects.ai/p/gemma-4-and-what-makes-an-open-model)

---
_Generated by EarlyTerms · https://earlyterms.com/term/gemma-4_