A team of developers at Hugging Face, an AI development platform, including co-founder and chief scientist Thomas Wolf, claim to have created an “open” alternative to OpenAI’s advanced research tool.
OpenAI introduced its deep research tool during a recent event, designed to scour the internet and compile research reports on a variety of topics. While the capabilities are impressive, access to deep research is currently restricted to users with a subscription to OpenAI’s $200-a-month ChatGPT Pro plan.
The Hugging Face project, named Open Deep Research, integrates an AI model known as OpenAI’s o1 along with an open-source “agentic framework” designed to assist the model in planning its analyses and effectively utilizing tools like search engines. Although o1 is a proprietary model requiring a paid API for access, the developers assert it outperforms “open” models such as DeepSeek’s R1.
In under 24 hours, the researchers managed to employ o1 to operate a simplistic, text-based browser and a “text inspector” toolkit, allowing it to access files throughout the web. According to the team, Open Deep Research can autonomously browse the internet, navigate web pages, manipulate files, and execute calculations based on the data it encounters.
On GAIA, which serves as a benchmark for general AI assistants, Open Deep Research has achieved a score of 54%. This is in comparison to the 67.36% score recorded by OpenAI’s deep research tool.
I attempted to use Open Deep Research via the public demo provided by the team, but encountered difficulties. The site was heavily congested at the time of writing, and after waiting for 10 minutes, I received an error message.
Nonetheless, the research team is dedicated to enhancing the user experience and has made the source code available on GitHub for public review and feedback.
It’s important to note that there are numerous reproductions of OpenAI deep research available online, many of which utilize open models and tools. However, both these reproductions and Open Deep Research lack access to the o3 model that forms the core of deep research.
To date, few models can compete with o3 regarding benchmarks that involve answering intricate questions and gathering information. Without a viable open model to match o3, alternatives to deep research may fall short of providing an equivalent experience.
Subscribe to TechCrunch’s AI-focused newsletter! Sign up here to receive it every Wednesday in your inbox.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
