Hugging Face just pulled off something amazing—they created Open Deep Research, an AI that can browse the internet and generate research reports automatically, in just 24 hours! This is an open-source version of OpenAI’s Deep Research, which was previously only available commercially.
AI Technology Race
Here’s the deal—OpenAI recently launched Deep Research, a feature that collects information from various sources and compiles it into a report. But Hugging Face didn’t want to be left behind. They took on the challenge of replicating the system in just one day while making it free and open for all developers!
“Advanced Large Language Models (LLMs) are already available as open-source, but OpenAI hasn’t shared much about the agent framework behind Deep Research,” Hugging Face wrote in their announcement. “So we decided to take on a 24-hour mission to replicate their results and open-source the necessary framework along the way!”
How Good Is Hugging Face’s AI?
Even though it was developed in just one day, Open Deep Research is already showing impressive performance. In GAIA (General AI Assistants) testing, this AI reached an accuracy of 55.15%, while OpenAI’s Deep Research scored 67.36%. Not too far apart, right?
GAIA is a benchmark that evaluates AI’s ability to gather and organize information from multiple sources. For example, one of the toughest questions in GAIA challenges AI to identify which fruits appear in the 2008 painting “Embroidery from Uzbekistan” and whether those fruits were served in the breakfast menu of an ocean liner in October 1949. The answer has to be formatted in a specific way, listing the fruits in the order they appear in the painting. Sounds tricky? Even humans would struggle with that!
What Makes It So Cool?
The biggest advantage of Open Deep Research is its “agentic” system, which allows it to complete tasks step by step. The AI can browse the web, collect relevant data, and organize it into a full research report.
To build this AI, the Hugging Face team used large language models (LLMs) like OpenAI’s GPT-4o or reasoning models such as o1 and o3-mini. They also leveraged their own open-source library, “smolagents,” which controls AI actions using programming code instead of just JSON files. This makes the AI 30% more efficient at handling complex tasks.
What’s Next for Open Deep Research?
While it still needs some polishing to match OpenAI, Open Deep Research already provides a free alternative for developers who want to explore this technology. The Hugging Face team also plans to add new features like support for more file formats and even vision-based web browsing.
Interestingly, they’re also working on replicating OpenAI’s “Operator,” a tool that allows AI to control a mouse and keyboard inside a web browser. Looks like the AI competition is heating up!
For those interested, Open Deep Research’s code is available on GitHub, and Hugging Face is actively looking for AI engineers to help improve the project. The team is excited about the response from the community, saying it feels like “surfing a big wave” with so many contributors eager to join in.
What do you think? Can open-source AI like this become a serious competitor to tech giants like OpenAI and Google? Let us know your thoughts!