
This year is really shaping up to be the year of open-source artificial intelligence. We already had the very disruptive launch of the DeepSeek R1 LLM, which has given OpenAI's o1 a run for its money and now, Hugging Face is making big progress on a project called Open Deep Research, which aims to replicate the functionality of ChatGPT's Deep Research feature, using open source tools.
For those not familiar with OpenAI's Deep Research, it's a tool that takes a query you want to research, then goes out learning information from the web and synthesizes a well-researched report with plenty of citations. According to HuggingFace, Deep Research includes a large language model and an agentic framework that guides the LLM in using tools such as web search and organizing its actions in steps.
To keep open-source language models competitive, Hugging Face ran a 24-hour “mission” to try to replicate OpenAI's Deep Research and yielded several results. The community has created Open Deep Research, which performs quite competitively with Deep Research, but more work needs to be done, Hugging Face admitted.
In the GAIA validation sets, the Open Deep Research project scored 55.15%, while OpenAI's Deep Research scored a more impressive 67.36%. Interestingly, Hugging Face saw the best performance when using a coding agent compared to a regular agent that uses JSON. When using the standard agent, the score on the GAIA test fell to just 33%.
Hugging Face suspects that OpenAI's Deep Research is getting a performance boost from Operator, a recently announced agent that can browse the web and interact with web pages like a human. To try to fill this gap, Hugging Face wants to build its own web browser agent and has said it will be hiring a full-time engineer to help work on this problem. It also calls on anyone interested to help with the project voluntarily.
Hugging Face has released a live demo of Open Deep Research, but during our testing, there seemed to be an issue with tokens being exceeded, so we couldn't try it out.
4 Comments - Add comment