In a remarkable feat, Hugging Face researchers have developed an open-source AI research agent called “Open Deep Research” in just 24 hours. This project was initiated as a challenge following OpenAI’s launch of its Deep Research feature, which autonomously browses the web and generates research reports.
The discussion about Deep Research shows the gaps between fields for what they think “research” is.
OpenAI Deep Research is not great at digging up tons of citations or facts (Google’s is better at that)
Deep Research is great at digging up arguments, holes & providing analysis pic.twitter.com/w4PC2P5Jim
— Ethan Mollick (@emollick) February 6, 2025
Hugging Face aims to match the performance of Deep Research while making the technology freely available to developers.
Look, your experience may vary, but asking OpenAI's Deep Research about topics I am writing papers on has been incredibly fruitful. It is excellent at identifying promising threads & work in other fields, and does great work synthesizing theories & major trends in the literature.
— Ethan Mollick (@emollick) February 5, 2025
The company announced, “While powerful large language models (LLMs) are now freely available as open-source, OpenAI didn’t disclose much about the agentic framework underlying Deep Research. So we embarked on a 24-hour mission to reproduce their results and open-source the needed framework along the way!”
A great use case for OpenAI Deep Research is a 1-stop daily news report.
Prompt it with:
– General rules
– Personal bio
– Your interests
– Preferred sourcesIt’ll generate a comprehensive news report 100% customized to you.
This is how I’ll get my news now.
Full prompt below. pic.twitter.com/ANMIpTZEeA
— Mckay Wrigley (@mckaywrigley) February 3, 2025
Like OpenAI’s and Google’s “Deep Research” implementations, Hugging Face’s solution incorporates an “agent” framework to an existing AI model, enabling it to perform multi-step tasks like collecting information and dynamically building reports. Even after just one day of development, Hugging Face’s Open Deep Research has achieved 55.15 percent accuracy on the GAIA benchmark, which tests an AI model’s ability to gather and synthesize information from multiple sources.
OpenAI Deep Research is a dev supertool.
Ask it something you need docs for.
Take its analysis + codebase context and use o3-mini-high to condense it down to a plan.
Pass plan to o1-pro for an XML diff.
Tidy up loose ends in Cursor.
AI + code is literally a cheat code.
— Mckay Wrigley (@mckaywrigley) February 3, 2025
OpenAI’s Deep Research scored 67.36 percent on the same benchmark, with improved accuracy up to 72.57 percent when multiple responses were combined. GAIA includes complex questions demanding synthesis from various sources, making it a challenging test for AI. For instance, one question asks which fruits in a 2008 painting were served as part of an October 1949 breakfast menu on an ocean liner later used in a film, requiring detailed and contextual information retrieval.
An AI agent is fundamentally dependent on the underlying AI model. Open Deep Research currently operates on OpenAI’s large language models but is designed to be adaptable to open-weight AI models.
Open-source AI research agent unveiled
The agentic architecture is innovative, allowing the AI to complete research tasks autonomously. According to Hugging Face, building the right agentic layer is crucial, significantly enhancing the capabilities of large language models. The rapid development highlights the efficiency of open-source AI projects.
Open Deep Research leverages Hugging Face’s existing tools, such as their open-source library, making task sequences more concise and efficient. The open-source community’s involvement has been instrumental in advancing the project, with contributors building upon prior work, such as tools from Microsoft Research’s agent project from late 2024. Although Open Deep Research does not yet match OpenAI’s performance, its open availability invites developers to study and refine the technology.
Hugging Face’s project lead noted, “I think the benchmarks are pretty indicative. But in terms of speed and user experience, our solution is not as optimized as theirs.
Future improvements might include support for more file formats and vision-based web browsing capabilities. Hugging Face has shared the project on GitHub and opened positions for engineers to help expand its capabilities.
The project lead said, “The response has been great. We’ve got lots of new contributors proposing additions. It feels like catching the wave while surfing; the community really provides strong momentum.”
Hugging Face’s rapid and open-source replication of OpenAI’s Deep Research underscores the collaborative and innovative spirit driving the AI research community forward.
Rashan is a seasoned technology journalist and visionary leader serving as the Editor-in-Chief of DevX.com, a leading online publication focused on software development, programming languages, and emerging technologies. With his deep expertise in the tech industry and her passion for empowering developers, Rashan has transformed DevX.com into a vibrant hub of knowledge and innovation. Reach out to Rashan at [email protected]
