inclusionAI open-sourced Ming-3.0-asr-tiny, a context-aware ASR model for Chinese and English
inclusionAI released Ming-3.0-asr-tiny on Hugging Face, a context-aware speech recognition model for Chinese, English, and dialects, under the MIT License. The repository includes model weights, tokenizer, custom Transformers implementation, and a runnable inference example. The source release was published on 2026-10-14.
China context
- Original name
- 蚂蚁集团
- Outside China
- Open weights · huggingface.co
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can download the model weights and run inference locally using the provided PyTorch/Transformers example, enabling integration into applications without relying on a Chinese API.
- For investors
- The MIT-licensed release of a dialect-capable ASR model by an Ant Group-affiliated entity may pressure commercial ASR providers and expand the addressable market for speech-enabled products in Chinese-speaking regions.
Translated from Chinese. Quotes and facts link to the original sources.
inclusionAI released Ming-3.0-asr-tiny on Hugging Face. The model combines the Ling-tiny-v3 language backbone with an audio encoder to transcribe single-turn Chinese and English speech, cover regional dialects, and use hotwords or preceding dialogue to improve recognition of domain-specific terms. The repository includes model weights, tokenizer, custom Transformers implementation, and a runnable inference example. The model is evaluated on 15 regional dialect test sets, including Cantonese, Minnan, and Shanghainese, as well as noisy speech and five application domains. The model weights and original project code are released under the MIT License.
Ming-3.0-asr-tiny uses a Ling-tiny-v3 language backbone with an audio encoder. It supports context ASR by appending topic and hotwords to the base instruction, and can use dialogue context in the prompt. The example uses CUDA, BF16, SDPA, greedy decoding, and up to 256 new tokens. Dependencies include Transformers 5.12.1, fla-core 0.5.2, and PyTorch 2.11.0. The model does not provide word-level timestamps or speaker diarization, and the included example does not support streaming input.
Developers outside China can now use a MIT-licensed ASR model that handles Chinese dialects and context-aware transcription, reducing the cost of building speech applications for Chinese-speaking markets without relying on proprietary APIs.
The MIT license allows commercial use, modification, and redistribution, enabling businesses to integrate Chinese and English ASR with dialect support into products without licensing fees. The context ASR feature can improve accuracy for domain-specific terms, potentially reducing post-processing costs.
Next signals to check: whether inclusionAI releases larger Ming-3.0 ASR variants, whether the model is integrated into popular ASR pipelines, and whether independent benchmarks confirm the reported dialect and domain performance.