OpenAI has launched a new artificial intelligence tool called Deep Research, which Deutsche Bank analysts say outperforms all other AI models in answering complex academic and industry questions.
It said the new artificial intelligence tool scored 26.6% on the “Humanity’s Last Exam” (HLE), a new test designed to assess AI capabilities across more than 1,000 subjects.
By comparison, OpenAI’s previous model, GPT-4o, scored just 3.3%. The underlying o3 language model also outperformed human PhD researchers using Google in their own fields, according to a leading benchmark called GPQA Diamond.
Deep Research is designed to generate detailed and well-sourced reports in record time.
Deutsche Bank tested the system by asking it to analyse the impact of new US steel and aluminium tariffs.
In just eight minutes, it produced a 9,000-word report citing 22 sources, a level of depth and speed far beyond what previous AI models could achieve.
Analysts noted that while the responses weren’t flawless, they showed significant improvements in clarity, relevance, and accuracy.
The system’s ability to prioritise credible sources and cite them directly also appeared to reduce the risk of AI-generated misinformation, or "hallucinations."
At launch, Deep Research costs $200 per month, but OpenAI CEO Sam Altman has suggested a free version could be made available in the future.
Deutsche Bank describes the tool as a potential game-changer, offering businesses and researchers the equivalent of an expert-level PhD assistant at a fraction of the cost of hiring a human expert.