The directory
Every AI coding harness we track, one page each: what it costs, what it runs on, what it measures. Stars are a proxy for mindshare, not quality. Token cost is measured on our fixed 10-task suite where marked; everything else is observed from the project itself.
| Harness | Stars | Language | License | BYOK | Tokens (our suite) |
|---|---|---|---|---|---|
| OpenCode | 209K | TypeScript | MIT | Yes | 10.0M |
| Claw Code | 195K | Rust | Open source | Yes | n/a |
| Claude Code | 148K | TypeScript | Proprietary | No | 23.1M |
| Codex CLI | 126K | Rust | Apache-2.0 | No | n/a |
| Pi | 109K | TypeScript | MIT | Yes | 7.2M |
| Gemini CLI | 107K | TypeScript | Apache-2.0 | No | n/a |
| OpenHands | 89K | TypeScript | MIT | Yes | n/a |
| Cline CLI | 69K | TypeScript | Apache-2.0 | Yes | n/a |
| Goose | 54K | Rust | Apache-2.0 | Yes | n/a |
| Aider | 49K | Python | Apache-2.0 | Yes | n/a |
| Codewhale | 41K | Rust | Open source | Yes | n/a |
| 3code | 86 | Nim | MIT | Yes | 5.1M |
Token figures: 10-task SWE-bench Verified subset, same LiteLLM proxy, same provider, vanilla configs — methodology. Listings are refreshed monthly. This site is run by the maintainer of one listed harness (3code); every claim is sourced and the raw benchmark data is public.