Deconstructing Youdao's 'Zhi Yue 4' Fully Open Source: How to Reduce Implementation Costs Through Refactored Chain-of-Thought?
NetEase Youdao has announced a significant upgrade to its "Confucius 4" large language model, now entering the multimodal era with support for text, image, and audio interactions. The company is open-sourcing its core multimodal and text-to-speech (TTS) models, aiming to reduce implementation costs for developers. The new model demonstrates state-of-the-art performance in visual mathematical reasoning and offers a 43.2% reduction in reasoning chain output length, leading to lower inference costs. AI
IMPACT Lowers barriers for developers in multimodal and speech synthesis, potentially accelerating AI Agent development and adoption.