Quant-Arena (iterative total score)
Professional Work · 2026-10-21
End-to-end quantitative trading research on real China A-share data: a model iteratively builds features and stock-scoring code with feedback only on 2022-2023, then freezes one strategy for a sealed 2024 test in a cash-account simulator with trading costs and market impact. The iterative-track total score (0-100) weights the sealed period 0.7 and the research period 0.3; each combines realized excess return (dominant), Sharpe, signal quality and feature diversity.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 51.3 |
| GPT-6 Sol | 51.1 |
| GPT-5.6 Terra | 47.6 |
| Claude Opus 5.5 | 44.3 |
| Gemini 3.8 Flash | 43.0 |
| Claude Sonnet 5.5 | 40.9 |
| GLM-5.3 | 38.9 |
| Kimi K3 | 38.3 |
| DeepSeek V4.1 Flash | 36.5 |