An open-source 30B MoE model rivaling OpenAI DeepResearch
About
Tongyi DeepResearch is the first fully open-source web agent to perform on par with OpenAI DeepResearch, achieving state-of-the-art results on complex information-seeking benchmarks
Tongyi Lab is one of Alibaba Group's AI research and development divisions, focusing on large language models (LLMs), multimodal models, and agent technology; it created the QWEN models
Built an end-to-end training pipeline that integrates Agentic Continual Pre-training (CPT), Supervised Fine-Tuning (SFT), and Reinforcement Learning (RL)
Strengthens long-horizon reasoning and planning through large-scale QA generation based entirely on synthetic data and the IterResearch paradigm
Supports ReAct mode and Heavy mode to handle everything from simple reasoning to complex multi-step research, with stable performance secured through reinforcement learning based on the GRPO algorithm
Already deployed in services inside and outside Alibaba, such as Gaode Mate and Tongyi FaRui, demonstrating the practicality and scalability of open-source AI research agents