PANews reported on March 18 that AI company Sahara AI announced a collaboration with Microsoft, providing high-precision labeled data to jointly launch the open-source benchmark MATHVISTA. This benchmark is specifically designed to test the reasoning and decision-making capabilities of models such as GPT-4V, Claude, and Gemini in real-world scenarios, and has already been downloaded over 270,000 times. This type of high-quality labeled data is the foundation for AI agents to have reliable reasoning and decision-making capabilities, directly impacting the performance of agents used by millions of users daily. Currently, institutions such as Microsoft, Amazon, Snap, and MIT have adopted Sahara AI's data services and agentic AI solutions.