Gemma 4 overview and key facts (October 2026)

Gemma 4 is Google DeepMind's family of open-weight multimodal models in five sizes, with the smallest versions designed to run on laptops and mobile devices and the larger ones on workstations.

Gemma 4 comes in sizes E2B, E4B, 12B, a 26B mixture-of-experts model with 4 billion active parameters and 31B. All handle text and images, and E2B, E4B and 12B also take audio. The small models have 128,000 tokens of context and the medium ones 256,000. All models support configurable thinking, native function calling and system prompts. The E4B version had more than 4.4 million downloads in the last month at our check, and the 26B version more than 12 million.

• Runs AI on devices without cloud accessE2B and E4B are designed for laptops and phones. Field staff get an assistant that works offline and keeps data on the device.
• Transcribes and understands audioNative audio input on the small models covers voice notes and short recordings. Apps can process spoken input locally.
• Builds agents on a workstationFunction calling and the 26B or 31B models support agent workflows on a single GPU workstation. Teams prototype without cloud costs.

• Five sizes from mobile to workstation under one Apache 2.0 license.
• Audio input on the small models, which is rare among open models of this size.

Gemma is built by Google DeepMind; Google LLC is based in Mountain View.

Self-hosting means your team runs, secures and updates the model, including GPU capacity, monitoring and patching. Check the license terms for commercial use and keep a record of which version you deploy. If outputs feed decisions about people, a human must stay in the loop. On-device deployment still needs a plan for updates across all devices.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews