88Z AI, 22 May — NetEase Youdao's AI team has open-sourced Confucius4, a multimodal reasoning model based on the Qwen3.5-27B architecture. Designed for advanced mathematical reasoning, it supports mixed text-image inputs and is compatible with the Qwen ecosystem.
At a glance
- Built on Qwen3.5-27B architecture for advanced math reasoning, reducing reasoning chain length by 43.2%
- Supports mixed text-image inputs and Chinese scenario optimization, achieving SOTA performance while compatible with the Qwen ecosystem
The update
NetEase Youdao's AI team has officially open-sourced Confucius4, a multimodal reasoning model. Built on the Qwen3.5-27B architecture, it is specifically optimized for advanced mathematical reasoning tasks. The model employs a training strategy combining iterative supervised fine-tuning with reinforcement learning, alongside refined chain-of-thought optimization. Benchmark results indicate that Confucius4 achieves state-of-the-art performance for its size on tests such as Math-Hard-500, while reducing reasoning chain length by 43.2%. Furthermore, the model is directionally optimized for Chinese user scenarios, supports mixed text-image inputs, and is fully compatible with the Qwen ecosystem, offering developers an efficient multimodal reasoning solution.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- Confucius4Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.