PolyBench - Advanced Property Reasoning: leaderboard
Open-ended polymer reasoning from PolyBench (Mohanty et al., Stony Brook University, 2026): explaining the mechanistic origins of properties, inferring properties from structural components and weighing trade-offs between several properties, scored 1-7 by a GPT-OSS-20B judge. Higher is better. Models suffixed PolyBench are the authors' fine-tunes on the benchmark's training split.
Metric: LLM-Judge Score (1-7 Likert; GPT-OSS-20B judge, open-ended answers). Source: arxiv.org. Saturation forecast: Around October 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 5.68 | #121 |
| 2 | Phi-4 | 5.19 | #1920 |
| 3 | Claude Sonnet 4.6 | 4.88 | #158 |
| 4 | GPT-4o | 4.73 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=polybench-advanced-property-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.