NetEase Youdao Open-Sources Ziyue 4 Multimodal and Text-to-Speech Models

网易有道"子曰4"多模态模型、语音合成模型全量开源

NetEase Youdao open-sourced its Ziyue 4.0 multimodal and text-to-speech models, with the 27B multimodal model reporting 81.4% accuracy on Chinese math reasoning tasks and the speech model supporting 14 languages.

Why it matters: HKR-H/K/R pass: the story has a concrete open-source hook, specific model numbers, and practitioner relevance. NetEase Youdao is not a frontier lab, so it stays below the 78+ good-quality band.