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

2026-04-02 · Source: Google DeepMind News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, long

Summary

Google has introduced Gemma 4, its latest family of open large language models, designed for advanced reasoning and agentic workflows. These models, built on the same technology as Gemini 3, offer unprecedented intelligence-per-parameter and are released under an Apache 2.0 license. Gemma 4 comes in four sizes: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B Dense. The 31B model ranks as the #3 open model globally on the Arena AI text leaderboard, with the 26B model at #6, outperforming models 20 times their size. The E2B and E4B models are optimized for mobile and IoT devices, supporting multimodal capabilities and low-latency processing. The larger models feature context windows up to 256K and native support for function-calling, structured JSON output, code generation, and multimodal input across 140+ languages.

Key takeaway

For AI Architects and NLP Engineers evaluating open-source models for deployment, Gemma 4 presents a compelling option due to its high intelligence-per-parameter and Apache 2.0 license. Your teams can leverage its advanced reasoning, agentic workflow support, and multimodal capabilities for efficient development on diverse hardware, from mobile devices to cloud infrastructure. Consider prototyping with Gemma 4 to reduce hardware overhead and enhance application functionality.

Key insights

Gemma 4 offers highly capable, open-source AI models optimized for diverse hardware, from edge devices to data centers.

Principles

• Maximize intelligence-per-parameter for efficiency.
• Prioritize open access and developer flexibility.
• Ensure multimodal and multi-language capabilities.

Method

Gemma 4 models are developed from Gemini 3 technology, released in various sizes (2B, 4B, 26B MoE, 31B Dense) under an Apache 2.0 license, and optimized for efficient fine-tuning and deployment across diverse hardware.

In practice

• Utilize Gemma 4 for local AI code assistants.
• Develop autonomous agents with native function-calling.
• Fine-tune models for specific tasks on various GPUs.

Topics

• Gemma 4
• Open Models
• Agentic Workflows
• Multimodal AI
• Apache 2.0 License

Code references

• NVIDIA-NeMo/Automodel

Best for: AI Architect, NLP Engineer, CTO, AI Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• Gemma 4: Byte for byte, the most capable open models — Gemma 4 offers unprecedented intelligence-per-parameter in open models, enabling advanced reasoning and agentic workflows across diverse hardware.
• [AINews] Gemma 4: The best small Multimodal Open Models, dramatically better than Gemma 3 in every way — Gemma 4 advances open-source AI with multimodal capabilities, efficient architecture, and a permissive Apache 2.0 license.
• Gemma, DeepMind's Family of Open Models — Omar Sanseviero, Google DeepMind — Gemma 4 offers highly capable, open, and on-device LLMs with an Apache 2 license and multimodal multilingual support.
• What’s new in Gemma 4? — Gemma 4 offers a family of open-source, locally runnable models for diverse hardware, from mobile to desktop.
• Gemma 4 Local Test | New Open LLM King? — Gemma 4 offers strong multimodal capabilities and agentic workflow support under an Apache 2 license.
• Gemma 4 — Gemma 4 offers powerful, open-weight multimodal AI with strong local deployment and agentic capabilities under an Apache 2.0 license.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Google DeepMind News.