LilyBench - Zero-Shot Compile Rate: leaderboard
LilyBench (2026) tests language models on symbolic music written in LilyPond notation. This board is the percent of zero-shot generations from a 200-prompt bank of Baroque-style requests that the LilyPond binary compiles to MIDI; higher is better.
Metric: Compile rate (%) of zero-shot LilyPond generations over a 200-prompt stratified bank (natural-language instruction plus composer, period, form, ensemble and part metadata); a generation counts when the LilyPond binary renders it to MIDI; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Phi-4 | 71.1 | #1913 |
| 2 | Qwen 2.5 Coder 14B | 69 | #3314 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=lilybench-zero-shot-compile-rate · How It Works · Data refreshed daily, snapshot 2026-11-13.