• 34ºc, Sunny

OpenAI Launches New Agent Deep Research Outstanding in "Humanity's Final Exam"

OpenAI has released a new artificial intelligence tool, "Deep Research", designed to perform time-consuming online research tasks for users, covering a wide range of fields from complex scientific problems to car recommendations. This new tool further enriches the startup's line-up of artificial intelligence agents. Deep Research can take anywhere from five to 30 minutes to do its job - digging deep into the network while you're away or tackling other tasks. The final output appears in the form of a report in the chat, and soon users will also see embedded images, data lake visualization, and other analytical output for enhanced clarity. This time the time has greatly improved its capabilities. Deep Research, for example, achieved an impressive 26.6% accuracy on the "Final Exam for Humanity" - well ahead of its closest competitor, DeepSeek R1, which scored under 10%. The leap highlights the system's iterative optimization and structured general capabilities, proving that some tasks do take time at the rapid pace of artificial intelligence.