Hugging Face is developing an open version of OpenAI's AI agent 'Deep research'



It has been discovered that developers at the AI development platform Hugging Face are building an open version of '

Deep research ,' an AI agent announced by OpenAI that searches for information on the internet and makes inferences.

Open-source DeepResearch – Freeing our search agents
https://huggingface.co/blog/open-deep-research




Hugging Face researchers aim to build an 'open' version of OpenAI's deep research tool | TechCrunch
https://techcrunch.com/2025/02/04/hugging-face-researchers-aim-to-build-an-open-version-of-openais-deep-research-tool/

Deep research is an AI agent that can search for information on the Internet and has three main functions: 'content discovery,' 'content integration,' and 'information reasoning.' Deep research will be integrated into ChatGPT and made generally available, and is already available in the subscription service ChatGPT Pro.

ChatGPT was not able to search for information on the Internet and could only respond to user requests using information given to it in advance, but it is expected that the integration of Deep research will enable it to handle user requests more flexibly.

OpenAI announces that ChatGPT will have a 'Deep research' feature that allows it to collect online information - GIGAZINE



However, at the time of writing, Deep Research is only available to paid members, and all that is known about the future roadmap is that it will be gradually rolled out to various paid members.

In this situation, it was discovered that a group of developers including Hugging Face co-founder and chief scientist Thomas Wolf are building an 'open' version of Deep Research called 'Open Deep Research.'

According to Wolf et al., the deep research of the original consists of 'large-scale language models (OpenAI 4o, o1, o3, etc.)' and 'agent frameworks'. The agent framework is a tool built into the AI system and is used by the LLM to browse websites and read PDF files.

Noting that OpenAI has not made much public about the agent framework underlying its deep research, Wolf and his colleagues decided to create their own open source agent framework.

Open Deep Research consists of OpenAI's 'o1' and an open source agent framework. o1 is a proprietary model, that is, it is protected by a paid API, but the agent framework is open. Wolf and his colleagues say that by developing the agent framework as open source, users will be able to use their favorite models, including 'o1', and use fully locally customized AI agents.

First, Wolf and his team built a system that uses a simple text-based browser and the Text Inspector toolkit to read files on the web, aiming to achieve performance similar to that of Deep Research. This was achieved within 24 hours of Deep Research being published.

According to Wolf et al., in the benchmark ' GAIA ', where Deep Research recorded an average score of 67.36%, Open Deep Research achieved a score of 54%. Although the difference is clear, Wolf et al. stated that 'this is just the beginning, and there is much to improve,' indicating their intention to continue development.

You can access the demo version of Open Deep Research from the URL below, and get Open Deep Research's answer by entering the prompt in the section that says 'Your request' and pressing the enter key.

open Deep Research - free the AI agents!
https://m-ric-open-deep-research.hf.space/



However, when I actually tried it, I found that there were more than 90 queues from users around the world, and I had to wait 11,394 seconds (about 3 hours) for an answer to be generated.



But Wolf and his team are committed to improving the experience, and have even made the source code available on GitHub for viewing and feedback.

smolagents/examples/open_deep_research at main · huggingface/smolagents · GitHub

https://github.com/huggingface/smolagents/tree/main/examples/open_deep_research



Related Posts:

in Software, Posted by log1p_kr