DeepSeek and OpenAI: The Growing AI Rivalry
Since DeepSeek challenged OpenAI two weeks ago, the open- vs. closed-source AI competition has shown no signs of stopping. This dynamic landscape is shaping the future of artificial intelligence, bringing different philosophies and technological advancements to the forefront.
OpenAI's Deep Research vs. Hugging Face's Open Deep Research
Just two days after OpenAI announced Deep Research, a new AI agent within ChatGPT designed to sift through online sources, its open-source counterpart emerged. Hugging Face introduced its version of this feature, known as open Deep Research, which serves as a significant milestone in the open AI community.
Performance Comparison
Hugging Face's open Deep Research uses OpenAI's o1 model and an agentic framework to navigate the web. According to the results, the open alternative achieved 55% accuracy on the General AI Assistants benchmark (GAIA), a prominent assessment test for AI agents. In comparison, OpenAI's Deep Research attained a 67% accuracy, indicating a notable lead but also highlighting the progress made by open-source initiatives.
| AI Agent | Platform | Accuracy (GAIA) |
|---|---|---|
| Deep Research | OpenAI | 67% |
| Open Deep Research | Hugging Face | 55% |
Hugging Face acknowledges that their agent is still not a full competitor to OpenAI's. In their blog titled "Freeing Our Search Agents," they state, "Deep Research is a massive achievement, and its open reproduction will take time. Full parity will require improved browser usage and interaction beyond the current text-only web interaction we explore in this first step."
The Advanced o3 Model
OpenAI's Deep Research is built on the latest and most advanced reasoning model, o3, which currently has no known open-source equivalent. According to OpenAI's blog, this model outperformed other top models on Humanity's Last Exam (HLE), a new AI benchmark test that is much more challenging than other popular tests. The o3 model achieved nearly 27% accuracy, "a new high" for this test.
However, HLE's creators have pointed out a potential "contamination": the o3 model was evaluated after HLE's release, meaning OpenAI had access to its prompts. Hugging Face has not disclosed whether it tested open Deep Research on HLE.
Future Developments and Accessibility
To better compete with proprietary models, Hugging Face plans to build "agents that view your screen and can act directly with mouse and keyboard." This development aims to bridge the gap between open-source agents and their proprietary counterparts, offering more comprehensive interaction capabilities.
Considering Deep Research's $200-per-month price tag via ChatGPT Pro, it may be inaccessible to most users. However, Hugging Face offers open Deep Research's live demo for free, referring to it as a "simplified version" of the full agent. This allows users to experiment with the technology without financial commitment.
The rapid pace at which Hugging Face created a competitor—under 24 hours—underscores the intense race in the AI field. Researchers at UC Berkeley developed a model comparable to OpenAI's o1-preview in just 19 hours earlier last month. Although DeepSeek's exact timeline on its o1 rival model, R1, is unknown, it is understood to be lower in terms of resource requirements, both in time and expenditure.
Artificial Intelligence Innovations and Future Prospects
As the AI competition between open and closed-source models heats up, the technological advancements and rapid developments promise exciting prospects. Hugging Face, OpenAI, and other entities are pushing the boundaries, ensuring that end-users benefit from increasingly sophisticated and accessible AI tools.
Whether it's in coding, data analysis, or general AI assistance, the evolving landscape reveals a future where both proprietary and open-source models will play crucial roles in advancing the field of artificial intelligence.
