ShaderMatch: leaderboard
Shader code generation benchmark: evaluates code models on generating GLSL shader programs that visually match reference outputs across multiple clone-detection levels.
Metric: Clone Match Rate (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 49 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | starcoder2-15B | 18.78 | #1495 |
| 2 | deepseek-coder-6.7B-base | 17.51 | #1484 |
| 3 | CodeQwen1.5-7B | 16.75 | #1514 |
| 4 | Yi-Coder-9B | 14.72 | #1428 |
| 5 | CodeLlama-7B-hf | 13.71 | #1457 |
| 6 | starcoder2-7B | 10.15 | #1572 |
| 7 | Llama 3.1 8B | 9.9 | #1151 |
| 8 | Stable-Code 3B | 9.64 | #1595 |
| 9 | starcoder2-3B | 7.36 | #1596 |
| 10 | Phi-2 | 1.52 | #1511 |
| 11 | Phi-1 | 1.52 | #1639 |
| 12 | Phi-1.5 | 0.76 | #1549 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=shadermatch · How It Works · Data refreshed daily, snapshot 2026-10-20.