Whoa, OpenAI and Paradigm have jointly launched EVMbench. This is a benchmark specifically designed to evaluate AI agents' capabilities in the blockchain security space. The tool assesses models' ability to handle high-risk vulnerabilities in Ethereum Virtual Machine (EVM) smart contracts across three dimensions: detection, remediation, and exploitation. The latest GPT-5.3-Codex performed exceptionally well in exploitation tasks, showcasing significant progress in AI's ability to handle complex code logic.