claude code vs cursor vs codex — measured, not surveyed
| # | Agent | Devs | Tokens | Share | Top model | Est cost |
|---|---|---|---|---|---|---|
| 01 | Codex CLIcodex | 48 | 56.83B | 53% | gpt-5.6-sol | $46,954 |
| 02 | Claude Codeclaude | 60 | 41.37B | 39% | claude-opus-5 | $33,332 |
| 03 | OpenCodeopencode | 17 | 4.94B | 4.6% | deepseek-v4-flash-free | $317 |
| 04 | Cursorcursor | 19 | 3.70B | 3.4% | default | $3,053 |
| 05 | Hermeshermes | 7 | 258.9M | 0.2% | hy3:free | $289 |
| 06 | Kimi CLIkimi | 2 | 188.0M | 0.2% | k3 | $81.47 |
| 07 | Pipi | 2 | 129.5M | 0.1% | model-237bf97a0500 | $7.72 |
| 08 | Clinecline | 1 | 143.8K | <0.1% | glm-5.2 | $0.22 |
| 09 | Kilo Codekilo | 1 | 13.8K | <0.1% | free | $0 |
npx whoburnedmore CLI reads the local usage logs coding agents already keep and submits daily aggregate token totals per tool and model — never prompts, code or file names. Panel = 98 listed developers who opted in publicly, including subscription usage (Claude Code, Cursor, Codex) that API-gateway rankings never see. Costs estimated from public per-token pricing. what we collect · leaderboard · state of AI coding