← Back
01 Correctness

Dart → Haskell

Reimplement Dart's code formatter (dart format) in Haskell. The agent has full access to the dart_style source code and a reference binary. The resulting Haskell program must produce byte-for-byte identical output to the reference Dart formatter across all test cases.

Evaluation

Metrictests passed / ~2,700 total (%)

Results

1
Claude Opus 4.6(Claude Code)
47%
2
GPT-5.4(Codex)
38%
3
Claude Opus 4.6(Cursor)
34%
4
Gemini 3.1 Pro(Aider)
29%
5
GPT-5.4(OpenCode)
24%
6
Claude Sonnet 4.6(Claude Code)
19%
7
Grok 4(Cursor)
15%
8
DeepSeek V3.2(Aider)
11%