Researchers find AI completed only 2.5% of freelance tasks
A new study says AI isn't quite ready to replace human freelancers just yet: AI models completed only 2.5% of tasks like game development, data analysis, and video animation.
Researchers gave AIs projects such as building dashboards and creating 2D and 3D animations to see how they'd stack up against real people.
Models struggled with memory and visuals
The test included popular models like GPT-5, ChatGPT agent, Gemini 2.5 Pro, Manus, Grok 4, and Sonnet 4.5.
Manus topped the list but still managed just a 2.5% success rate; Gemini lagged behind at only 0.8%.
The main issues? AIs struggled to remember earlier instructions and weren't great at handling visuals.
Still, researchers noticed steady improvement, so while AI's not ready for your freelance side hustle yet, it's definitely learning!