On August 4, 2026, Tencent Hunyuan proudly introduced its latest-generation speech recognition model, the Hy ASR 3.0 preview. This innovative model seamlessly blends high-precision speech recognition capabilities with profound semantic comprehension, achieving remarkable enhancements across key aspects including general recognition accuracy, contextual awareness, adaptability to diverse scenarios, and dialect inclusivity. In open-source evaluation datasets, the Word Error Rate (WER) for various languages is maintained at approximately 3%, with specific rates of 3.34% for Mandarin Chinese, 2.62% for English, and 3.12% for Cantonese. Presently, the Hy ASR 3.0 preview is accessible through API services on the official Tencent Cloud website. Yuanbao has successfully completed its initial rollout, providing complimentary access to features like dialect recognition and contextual error correction. Meanwhile, products such as WorkBuddy are also undergoing integration to leverage this advanced technology.