Solving Is Not Drawing - Compile Success Rate (With Reference Solution): leaderboard
Solving Is Not Drawing (2026) asks models to draw the diagram of a self-contained olympiad geometry problem as Asymptote code, and compares the result with a human-authored reference diagram. The model receives the problem with its full reference solution and only has to render a consistent figure. Score is the share of generated programs that render without error, in percent; higher is better.
Metric: Compile success rate (%; share of generated Asymptote programs that render to an image without error, one program per problem, on the olympiad geometry benchmark (AMC, AIME, USAJMO, USAMO and IMO problems with human-authored Asymptote diagrams; the 297-problem hard subset rated 7 or more of 10); the model receives the problem statement with the full reference solution and approach). Source: arxiv.org. Saturation forecast: Around March 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Flash Lite | 59.2 | #364 |
| 2 | GLM-5.1 | 40 | #179 |
| 3 | Kimi K2.5 | 30.9 | #256 |
| 4 | Claude Sonnet 4.6 | 27.6 | #161 |
| 5 | GPT-5.4 Mini | 21.7 | #389 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=solving-is-not-drawing-compile-success-rate-with-reference-solution · How It Works · Data refreshed daily, snapshot 2026-11-13.