On July 6, 2026, the QianWen large model underwent a significant transformation, officially upgrading to the cutting-edge real-time speech recognition large model, Fun-ASR-Realtime. This streaming model boasts an impressive first-character latency that is meticulously controlled at the hundred-millisecond mark, ensuring swift and seamless recognition. Its accuracy rivals that of offline models, providing unparalleled precision in speech-to-text conversion. Additionally, it supports a vast array of 16 dialects and 30 languages, catering to a diverse range of users and covering an extensive 46 speech recognition scenarios.

During the Film & Television Tornado's marathon 100-hour live broadcast, the Fun-ASR-Realtime model proved its mettle by delivering real-time subtitles throughout the entire event. Its remarkable performance resulted in a cumulative recognition of a staggering 1.32 million words, showcasing its robustness and reliability in high-pressure, real-world applications.