# Multi-agent tasks: which coding agent

> Weekly ranking of four coding-agent CLIs (Claude Code, Codex CLI, Antigravity CLI, Grok CLI) for five task classes (coding, agentic, research, reading, writing), with coding and agentic work also ranked by kind (for example bug fixing, frontend, terminal work), from public leaderboards and own job outcomes.

## Data

- [bench.md](/bench.md): compact Markdown routing table; start here.
- [bench.json](/bench.json): full ranking with every scoring input and the method (schema_version 4).