GLM-ASR Nano 2512
zai-org/GLM-ASR-Nano-2512
published Dec 2025 · updated Apr 2026
GLM-ASR Nano 2512 is a robust, open-source automatic speech recognition (ASR) model optimized for dialect support and low-volume speech.
specs
| Task | Automatic Speech Recognition (ASR) |
| Architecture | Transformer-based seq2seq |
| Parameters | 1.5B |
| Supported Languages | 17 languages (WER ≤ 20%) |
about this model
GLM-ASR-Nano-2512 is a speech recognition model with 1.5 billion parameters that transcribes audio into text, optimized for Mandarin, English, Cantonese, and 14 other languages with WER ≤ 20%.
Key Capabilities
- Dialect support: Highly optimized for Cantonese and other dialects beyond standard Mandarin and English.
- Low-volume speech robustness: Trained to accurately transcribe extremely quiet or whispered audio that traditional models often miss.
Benchmark Performance
The model achieves the lowest average error rate (4.10) among comparable open-source models, outperforming OpenAI Whisper V3 on multiple benchmarks. It shows significant advantages in Chinese benchmarks including Wenet Meeting (real-world meeting scenarios with noise and overlapping speech) and Aishell-1 (standard Mandarin).
The model supports 17 languages with high usability and is available as a hosted, OpenAI-compatible API on gigarouter.
best for
- ·Transcribing Cantonese and other dialects
- ·Low-volume or quiet speech recognition
- ·Real-world meeting transcription with noise and overlapping speech
FAQ
It supports 17 languages with high usability, including Mandarin, English, Cantonese, and others.
It has 1.5 billion parameters.
It outperforms Whisper V3 on multiple benchmarks, achieving the lowest average error rate (4.10) among comparable open-source models.
It accepts audio as a URL or numpy array, processed via the AutoProcessor.
Use the OpenAI-compatible endpoint with your API key; refer to gigarouter documentation for details.
We're benchmarking and onboarding GLM-ASR Nano 2512 as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.