Gemma 4 E4B IT

Gemma 4 E4B IT sits inside Google DeepMind's Gemma 4 family of open-weight models released under the Apache 2.0 license, alongside pre-trained and instruction-tuned variants. The E4B size uses a Mixture-of-Experts design with around 4.5 billion effective parameters, sitting between the smaller E2B and the larger dense and MoE configurations in the lineup. This architecture choice, combined with a hybrid attention approach and Per-Layer Embeddings, is aimed at efficient on-device execution on laptops and phones without sacrificing multimodal capability.

The instruction-tuned variant is built for practical, interactive use: it accepts text, images, and audio natively while producing text, and it adds configurable thinking modes for step-by-step reasoning, native function calling for agentic workflows, and structured conversations through system-prompt support. The model is pretrained on more than 140 languages with out-of-the-box coverage for over 35, and the family targets long-context work, coding assistance, and reasoning-heavy assistants that can run locally. Its blend of open weights, multimodal input, and edge-friendly efficiency makes it a strong fit for developers building private, responsive applications on consumer hardware.