Tencent Hunyuan Hy ASR 3.0 preview is a speech recognition model from Tencent that understands context instead of just transcribing sounds. Released on August 4, 2026, it targets developers and product teams building call-center transcription, meeting notes, subtitles, and in-car voice features where meaning matters more than raw phonemes.
Core Features
- Built on the Hy3 MoE large language model, fusing acoustic recognition with semantic understanding.
- Context-aware error correction that resolves homophones using conversation history.
- Hot-word injection to boost accuracy on brand names, people, and industry terms.
- Reliable in noisy and whispered conditions, with coverage of 10 dialect regions and 20 sub-dialects.
- Reported word error rates of 3.34% for Mandarin, 2.62% for English, and 3.12% for Cantonese.
Use Cases
- Enterprise call centers and customer-service transcription with domain vocabulary.
- Meeting recording and subtitle generation for video and short-form content.
- On-device and in-car voice assistants needing dialect and noise tolerance.
Pricing
Free inside the Yuanbao app for all users. Enterprise developers access the commercial API through Tencent Cloud, billed per the standard Tencent Cloud speech recognition plan, with no separate list price published for the preview.
Our Take
Best for Chinese and multilingual teams that need transcription which actually understands meaning, not just phonemes. The trade-off is that it is a preview model with API access gated through Tencent Cloud rather than a standalone self-hosted release.
Browse more AI audio tools on aifreetool.site.
