Hugging Face Researchers Seek to Develop an ‘Open’ Alternative to OpenAI’s Deep Research Tool
A team of developers at the AI development platform Hugging Face, which includes co-founder and chief scientist Thomas Wolf, have created an “open” alternative to OpenAI’s deep research tool.
OpenAI introduced deep research at an event on Sunday, a tool that crawls the internet to gather research reports on various topics. While it is an impressive innovation, deep research is currently accessible only in a limited preview for users subscribed to OpenAI’s ChatGPT Pro plan, priced at $200 per month.
The Hugging Face project, dubbed Open Deep Research, features an AI model called OpenAI’s o1, alongside an open-source “agentic framework” designed to assist the model in planning its analysis and to utilize tools such as search engines. Although O1 is a proprietary model (meaning it requires a paid API), the team claims it outperforms “open” models like DeepSeek’s R1.
In under 24 hours, the researchers utilized o1 to deploy a simple, text-based browser and a “text inspector” toolkit, allowing it to read files available on the web. According to the team, Open Deep Research can autonomously navigate the internet, scrolling through pages, managing files, and even performing calculations with data.
Open Deep Research scored 54% on GAIA, a benchmark for general AI assistants, compared to OpenAI’s deep research score of 67.36%.
I attempted to use Open Deep Research in the public demonstration set up by the team but was unable to get it to function. The page experienced high traffic at the time of publication; after waiting for 10 minutes, I received an error message.
However, the researchers are dedicated to enhancing the user experience and have made the source code available on GitHub for public review and feedback.
It’s important to note that several reproductions of OpenAI deep research exist online, some of which utilize open models and tools. The key element that both these reproductions and Open Deep Research lack is o3, the model that forms the foundation of deep research.
Few, if any, models outperform o3 in benchmarks focused on answering complex queries and gathering information. Without a competitive open model against o3, alternatives to deep research may not fully stack up to the original.