PolyBench - Structural Understanding: leaderboard
Metric: LLM-Judge Score (1-7 Likert; GPT-OSS-20B judge, open-ended answers). Source: arxiv.org. Saturation forecast: Around May 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 5.1 |
| 2 | Claude Sonnet 4.6 | 4.74 |
| 3 | Phi-4 | 4.43 |
| 4 | GPT-4o | 3.71 |
Interactive version: theaggregate.ai/benchmark?slug=polybench-structural-understanding · How It Works · Data refreshed daily, snapshot 2026-09-25.