On December 10, 2025, Zhipu AI took the significant step of officially open-sourcing its GLM-ASR series of speech recognition models. Building on this series, the company also unveiled a desktop-based "Zhipu AI Input Method". The GLM-ASR-2512 model boasts an impressively low character error rate of just 0.0717. Meanwhile, the GLM-ASR-Nano-2512 model is designed to operate locally. The newly launched input method comes packed with a range of useful functions, including speech-to-text transcription, translation, and text rewriting capabilities.
