BlockRunAI Tests 10 Cutting-Edge AI Models, 7 Scammed in Solana Payments

Techub News reports that Solana's official account forwarded BlockRunAI's test report stating that the organization equipped 10 cutting-edge AI models with wallets containing real USDC to conduct payment performance benchmark tests on the Solana network. The tests were divided into two scenarios: financial research and adversarial markets, simulating the purchase of real data or facing fraudulent sellers. Key results show that given AI agents $100, the best performing models wasted between $25 and $38, while the worst models wasted between $56 and $71. Among the 10 models, 7 were scammed at least once by fraudulent sellers, 6 models paid twice for the same data, and only 4 models did not have any fund usage errors. When answering open-ended financial research questions, no model had an accuracy rate above 50%; GPT-6.1 Sol led with a 50% accuracy rate, with a cost of $0.468 per question.