Alibaba has entered the competitive world of AI research tools with the official launch of Tongyi DeepResearch, an open-source agentic model purpose-built to challenge OpenAI’s top-tier Deep Research platform.
Announced on September 18, 2025, the new model claims near-matching performance to OpenAI’s flagship while excelling in computational efficiency and accessibility for global developers.
The debut takes place as the tech giant seeks to democratize high-end AI research capabilities, making advanced agentic reasoning more accessible beyond the confines of closed-source giants.
This signals a new era in AI where innovation is increasingly driven by collaborative, open contributions rather than gated by proprietary systems.
How does Tongyi DeepResearch rival OpenAI’s flagship agent?
Tongyi DeepResearch targets the most challenging AI research tasks, directly comparing itself with OpenAI’s Deep Research agent. While OpenAI’s offering has set the standard by integrating advanced research automation into ChatGPT since February 2025, Alibaba’s alternative deploys a Mixture of Experts architecture with about 30.5 billion total parameters.
Crucially, only 3-3.3 billion are activated per token, maximizing throughput while maintaining top-tier reasoning accuracy.
The model’s open-source nature allows developers worldwide to audit, contribute, and expand its capabilities, addressing a long-standing call for more transparency and adaptability in AI research agents.
Community feedback on major AI platforms, like Hugging Face, has already been largely positive, praising the efficiency and flexibility Alibaba introduced.
Did you know?
Tongyi DeepResearch activates only about 10% of its parameters per task, significantly reducing energy use while matching the performance of much larger U.S. AI models.
What benchmarks did Tongyi DeepResearch outperform?
In rigorous testing, Tongyi DeepResearch bested OpenAI’s Deep Research on several respected benchmarks. On Humanity’s Last Exam, the agent posted a 32.9% accuracy rate, while OpenAI’s comparable score was 26.6%.
The agent performed impressively on BrowseComp, with 43.4% in English and 46.7% in Chinese. It also scored 75% on Xbench-DeepSearch, confirming robust results across both English and Chinese tasks.
The dramatic boost in efficiency further sets the model apart. Because it activates only certain parameters, Tongyi DeepResearch can process a lot of data quickly and use less computing power without losing much performance.
This model is designed for enterprise use cases and research environments where both speed and accuracy are essential for large-scale adoption.
How is Alibaba integrating Tongyi DeepResearch into real-world apps?
Alibaba has already started to leverage Tongyi DeepResearch beyond the academic lab. The agent is active in Amap, Alibaba’s popular navigation app, where it assists users with advanced trip planning using newly developed web retrieval and reasoning skills.
In the legal sector, the technology powers Tongyi FaRui, handling complex case law searches with verified citation features.
These integrations show how flexible the model is and that Alibaba wants to quickly use advanced AI by putting it into important consumer and professional applications.
As usage grows, Alibaba is expected to offer more cross-platform agentic features for industries from logistics to finance.
ALSO READ | Why Did Samsung Delay Its XR Headset and Tri-Fold Phone Launch?
What’s the broader context of Alibaba’s AI strategy?
The release of Tongyi DeepResearch is part of Alibaba’s wider ambition to lead in the global AI landscape. Alongside DeepResearch, the company unveiled several additional models, including Qwen3-Next-80B-A3B, with 80 billion parameters, and Qwen3-Max-Preview, which scales to a trillion.
The Qwen3-ASR-Flash model, supporting multilingual audio transcription, further cements Alibaba's multi-pronged AI development strategy.
This aggressive rollout is not just a response to OpenAI and Google DeepMind but also part of a broader bet on open innovation.
By providing open-source tools and showing strong benchmark results, Alibaba hopes to gather a global developer base that will rapidly improve and scale its AI agent offerings.
How is the global AI research agent race evolving?
The timing of Alibaba’s announcement points to intensifying rivalry in the AI research agent market. OpenAI’s Deep Research set the standard for multi-step web and data retrieval earlier this year, but the emergence of robust open-source competitors indicates a shift toward more democratized research tooling.
Google DeepMind and other US companies are also advancing, underscoring the sector's swift progress.
International collaboration, recent government policies, and high-stakes deals, such as the $42 billion US-UK Tech Prosperity Deal, are fueling new waves of investment in AI research infrastructure, particularly in nuclear, quantum, and agentic models.
With Tongyi DeepResearch openly available, developers have powerful new options as the field of AI research agents becomes ever more vibrant and globally distributed.
Looking ahead, Alibaba’s bet on open-source innovation could accelerate the closing of performance gaps with the world’s leading proprietary AI tools.
The next phase of the AI competition will likely hinge on community contributions, efficiency achievements, and real-world integration, giving both businesses and users unprecedented power to shape the technology’s future.
Comments (0)
Please sign in to leave a comment