The potential benefit of Cursor CLI (vs CC and Codex) is that you can easily between all major models (by Anthropic, OpenAI, xAI, as well as Kimi K3 and GLM 5.2). I found it useful when reviewing work - e.g. I implement using Opus then review using Sol, etc. Models by different providers tend to have different perspective on things and they can find different issues with the code.