# PolyBench

LLM forecasting and trading on Polymarket binary markets. Each point-in-time snapshot gives the model the order book and live news, and its trades are executed against the simulated order book.

## Facts

- Maker: Sichuan University
- Year: 2026
- Use case: Financial trading and portfolios
- Size: 38,666 binary markets across 4,997 events; 36,165 predictions from 7 models (2026-02-06 to 2026-02-12)
- Public split: Pipeline code on GitHub; dataset via a OneDrive link in the README
- Grading: Directional accuracy, Confidence-Weighted Return (CWR), APY and Sharpe from simulated order-book execution
- Licence: Not stated
- Submission: none
- Products allowed: Not stated
- Latest version: arXiv v1 (2026-04-03)
- What it does not cover: Equities and other asset markets; the evaluation window is one week
- Last checked: 1 Nov 2026
- Links: paper https://arxiv.org/abs/2604.14199, code https://github.com/PolyBench/PolyBench
- Directory page: https://benchmarks.finance/benchmarks/polybench/
- Leaderboard: https://benchmarks.finance/leaderboards/polybench/

## Results

### PolyBench: Confidence-Weighted Return (CWR) % (higher is better, 7 models)

| Rank | Model | Maker | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | MiMo-V2-Flash | Xiaomi | 17.6 | Confidence-Weighted Return (CWR) % | base lot $10, 2026-02-06 to 2026-02-12 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 2 | Gemini-3-Flash | Google | 6.2 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 3 | Trinity-Large | Not stated | -9.2 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 4 | deepseek-v3.2 | DeepSeek | -11.4 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 5 | Grok-4.1-Fast | xAI | -12.1 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 6 | GPT-OSS 120B | OpenAI | -17.8 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |
| 7 | MiniMax-M2.5 | MiniMax | -25.2 | Confidence-Weighted Return (CWR) % | base lot $10 | Reported by the authors | Paper Table 5 | https://arxiv.org/html/2604.14199v1 | 3 Apr 2026 |


Source: https://benchmarks.finance/benchmarks/polybench/

Created and maintained by Moneda.