Hugging Face Challenges OpenAI with Open Research Tool

Hugging Face Develops Open Alternative to OpenAI’s Deep Research

Developers at the AI development platform Hugging Face, including co-founder and chief scientist Thomas Wolf, have announced the creation of an “open” counterpart to OpenAI’s recently unveiled deep research tool.

OpenAI’s Deep Research: A Limited Preview

OpenAI’s deep research capability, showcased during a recent event, functions by systematically exploring the internet to generate comprehensive research reports on various topics. Currently, access to this tool is restricted to a limited preview for subscribers of the $200 monthly ChatGPT Pro plan.

Introducing Open Deep Research

The Hugging Face team’s project, named Open Deep Research, utilizes an AI model – OpenAI’s o1 – alongside an open-source “agentic framework.” This framework assists the model in planning its analytical process and directs it in utilizing tools such as search engines.

While o1 itself is a proprietary model requiring a paid API for access, the team reports that it demonstrated superior performance compared to openly available models like DeepSeek’s R1.

Rapid Development and Capabilities

Within a single day, the researchers successfully integrated o1 with a basic text-based browser and a “text inspector” toolkit, enabling it to access and read files from across the web.

Open Deep Research is designed to navigate the internet independently, performing actions like page scrolling, file manipulation, and data calculations.

Performance Benchmarks

On the GAIA benchmark, which assesses general AI assistants, Open Deep Research achieved a score of 54%. This contrasts with OpenAI deep research’s score of 67.36%.

Initial Testing and Public Access

Initial attempts to utilize the public demo of Open Deep Research encountered difficulties due to high server load. After a ten-minute wait, an error message was displayed.

The researchers have expressed their commitment to enhancing the user experience and have made the project’s source code publicly available on GitHub for review and contributions.

The Importance of the Underlying Model

Several OpenAI deep research “reproductions” are emerging online, many leveraging open models and tools. However, a critical component missing from these alternatives, including Open Deep Research, is o3 – the model that powers OpenAI’s deep research.

The Challenge of Model Comparison

Currently, few models can rival o3’s performance on benchmarks evaluating complex question answering and information retrieval. Without an equivalent open-source model, alternatives to deep research may not fully match its capabilities.

Note: TechCrunch offers an AI-focused newsletter. Subscribe here to receive it weekly.