Open model

Gemma 4

by Google DeepMind

Google DeepMind's family of open-weight multimodal models, whose smaller versions take audio input and are trained for multilingual speech recognition and speech translation.

Self-hosted

Official site Documentation

Every figure about Gemma 4

Each figure is a row of a research dataset. Open it for the exact words of its source or the exact request it was read from.

Speech Recognition Speed: Real-Time Factor Data

Klemen 2026, Gemma 3 27B correction20268 RTFx
Klemen 2026, Gemma 3 27B selection20268 RTFx

Open ASR Leaderboard: Speech Recognition Accuracy, Tracked

google/gemma-4-E4B-itread October 27, 202611.38 %

All open models and engines, compared

Every figure about Gemma 4

Speech Recognition Speed: Real-Time Factor Data

Open ASR Leaderboard: Speech Recognition Accuracy, Tracked

Other open models and engines