Overview
GLM-ASR-Nano-2512 is an open-source automatic speech recognition model designed to transcribe speech in complex real-world conditions. It takes audio as input and produces text. The official announcement is dated December 9, 2025. (zhipuai.cn)
Capabilities and uses
- The publisher describes it as a 1.5B-parameter model; its Whisper V3 audio encoder has additional parameters that are excluded from that figure.
- It supports Mandarin, English, Cantonese and other dialects, with particular attention to quiet or whispered speech. The announcement states that it supports 17 languages and describes recognition of custom keywords and rare words.
- Potential uses include speech transcription, voice input and locally deployed speech-recognition applications. The model can be used through the Transformers ecosystem.
- In the publisher's reported benchmark comparison, its average error rate was 4.10 versus 6.93 for Whisper V3. This is a result for the publisher's selected tests, not a guarantee for every language or recording. (zhipuai.cn)
Limitations
Transcriptions can contain errors and should be reviewed where accuracy matters. A project maintainer states that the model supports audio within 30 seconds; longer recordings require an appropriate segmentation strategy. The official inference example uses 30-second chunks. The announcement identifies the separately named cloud model GLM-ASR-2512 as the basis of Zhipu's AI input method; its service specifications should not automatically be attributed to the Nano model. (github.com)
Pricing
As of November 17, 2026, official hosted API pricing for this exact Nano model, including input and output charges, could not be verified in either CNY or USD. The international pricing page lists the differently named cloud model GLM-ASR-2512, so its price is not assigned to Nano. The domestic pricing page requires JavaScript and its price table could not be read. Self-hosting uses the published model weights but may incur infrastructure costs. (docs.z.ai)
Official sources
- Publisher announcement — the direct page request timed out; the announcement text was read from the search engine's indexed copy. (zhipuai.cn)
- Official model card and project repository. (huggingface.co)
- Pricing pages checked: domestic platform and international documentation. (open.bigmodel.cn)