Mistral Vibe for Code vs Claude Code vs Cursor vs Codex: Four Agents Scored on One Scaffold-to-PR Task

Four coding agents were compared on their ability to complete a real engineering task: scaffolding a feature across multiple files, generating and running tests, and opening a pull request. The agents tested were Mistral Vibe for Code, Claude Code, OpenAI Codex, and Cursor, each scored across five dimensions including scaffolding capability, test generation, pull request workflow, surface coverage, and cost and control options. Mistral Vibe for Code ranked first with 22 out of 25 points, followed by Claude Code and OpenAI Codex tied at 21 points, and Cursor at 16 points. The comparison reflects documented features and vendor specifications rather than single-run performance tests, with the source noting that benchmark suites across products are not directly comparable and vendor claims should be verified independently.