Gemma 4
Google DeepMind · April 2026
● activeOpen Sourcedecoder onlymultimodal
Parameters31B (A4B / E4B / E2B)
Context Window128K tokens
VariantsGemma 4, Gemma A4B, Gemma E4B, Gemma E2B, Gemma 4 12B
Why It Matters
Pushed the boundaries of what open models can do, with reasoning capabilities that rival much larger proprietary models. Edge variants enabled frontier-quality AI on mobile and embedded devices.
Description
Google's most capable open model family, featuring a hybrid architecture with configurable 'thinking' modes for reasoning tasks. Includes edge-optimized variants (E2B, E4B) small enough for mobile phones and IoT devices, alongside a 31B dense model for workstation-class performance. Released under the Apache 2.0 license for maximum flexibility. Later expanded with a 12B model capable of native audio input.
Notable Milestones
- ▸E2B and E4B edge models run on mobile phones and IoT devices
- ▸12B variant supports native audio input without an encoder
- ▸Apache 2.0 license enables unrestricted commercial use
Key Innovations
Multimodal
MultimodalProcessing multiple types of input (text, images, audio, video) in a single model.
Open Weight
Open WeightModel weights are publicly released but training data/code may not be. Enables fine-tuning but not full reproduction.
Distillation
DistillationTraining a smaller 'student' model to mimic a larger 'teacher' model, preserving capability at lower cost.
Reasoning
ReasoningStructured step-by-step problem solving, often using chain-of-thought or tree-of-thought approaches.