Google Releases Gemma 4: The Most Capable Open Models to Date

Google introduces Gemma 4 — the most intelligent open models to date. Purpose-built for advanced reasoning and agentic workflows, Gemma 4 delivers an unprecedented level of intelligence-per-parameter.

Model Sizes

Gemma 4 comes in four versatile sizes, moving beyond simple chat to handle complex logic and agentic workflows:

• E2B (Effective 2B) — Edge-optimized, 128K context window
• E4B (Effective 4B) — Edge-optimized with native audio input support
• 26B MoE — Mixture of Experts, activates only 3.8B parameters, latency-focused
• 31B Dense — Dense model maximizing raw quality, 256K context window

Core Capabilities

• Advanced Reasoning — Multi-step planning and deep logic, significant improvements in math and instruction-following
• Agentic Workflows — Native support for function-calling, structured JSON output, system instructions
• Code Generation — High-quality offline code, turning workstations into local AI coding assistants
• Multimodal — Native video and image processing, OCR and chart understanding; E2B/E4B support audio input
• Long Context — 128K for edge models, 256K for larger models, can process entire codebases
• 140+ Languages — Natively trained on over 140 languages

Performance

The 31B model ranks #3 among open models on Arena AI text leaderboard, with the 26B model at #6. Gemma 4 outcompetes models 20x its size.

Open Source License

Gemma 4 is released under Apache 2.0 license, providing complete developer flexibility and digital sovereignty for building and deploying across any environment.

Learn more: https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/