The rise of Chinese technology companies in the AI race continues. Recently, Alibaba unveiled its new AI model, a direct competitor to OpenAI’s advanced research model, Deep Research. Announced as “Tongyi DeepResearch,” the model is attracting attention not only for its robust performance but also for its accessibility. Alibaba has released its new AI tool as open source and free on the Hugging Face platform.
Tongyi DeepResearch Officially Launched
Tongyi DeepResearch is an AI model that, like Deep Research, focuses on multi-step research processes. Unlike traditional chat-based systems like ChatGPT, these models don’t just answer questions; they also collect real-time information from the internet, combine it, analyze it, and deliver more comprehensive output.
According to Alibaba’s statement, the model has 30 billion parameters. However, thanks to its Mixture of Experts (MoE) architecture, only 3 to 3.3 billion parameters are activated in each operation. This architectural choice provides the model with high speed and processing efficiency. Despite only using a small set of parameters, the model’s performance is competitive with much larger structures.
Tongyi DeepResearch is also notable for its success in various tests. In an advanced AI benchmark known as “Humanity’s Final Test,” it achieved a 32.9% accuracy rate, surpassing OpenAI’s Deep Research model’s 26.6% score.
It also achieved a 43.4% accuracy rate for English and 46.7% for Chinese in the BrowseComp test. These scores demonstrate the model’s strong multilingual support and content scanning capabilities.
With this model, Alibaba aims not only to gain a technical advantage but also to make AI more accessible. Open-source Tongyi DeepResearch makes an advanced tool freely available to both developers and researchers.
