Stepfun; Tsinghua University (with FinStep, Shanghai Jiao Tong University, Adelaide University), 2026
KTD-Fin
End-to-end LLM trading on CSI300 A-shares, with tickers and dates masked so models cannot recall history. A Barra-style attribution splits returns into market, style and stock-selection alpha.
About this benchmark
- Size
- 10 frontier LLM agents plus 18 Qlib factor baselines; 548-trading-day window (2024-01-01 to 2026-04-10) plus 10 regime windows; 4 masking levels x 3 decision modes
- Public split
- none found (paper says it is released as a template, but gives no URL)
- Grading
- Total return and Barra-attributed selection alpha (leaderboard ranks by alpha); reliability panel includes ECE; Brier score is recorded for diagnostics only
- Licence
- Not stated
- Submission
- none
- Products allowed
- Not stated
- Latest version
- arXiv v1 (2026-05-27)
- What it does not cover
- Markets outside Chinese A-shares, and inputs other than price data (observations are OHLCV only)
Results
10 modelsEach score shows who reported it: the benchmark's authors (quoted from their paper), the benchmark's official board, or Moneda (run with the benchmark's own grader and rules). We do not copy other aggregators.
KTD-Fin arXiv v1 (2026-05-27)
total return %, higher is better. 10 models.
Updated 27 May 2026
- Qwen3.6-Plus85.3
- GPT 5.561.3
- doubao-seed-2.061.0
- claude-opus-4-758.8
- minimax-m2.756.3
- step-3.5-flash21.9
- gemini-3.1-pro8.3
- deepseek-v4-pro1.4
- GLM-5.1-3.5
- Kimi K2.6-24.2
KTD-Fin arXiv v1 (2026-05-27)
total return %, higher is better. 10 models.
Updated 27 May 2026
- 1Qwen3.6-Plus85.3
- 2GPT 5.561.3
- 3doubao-seed-2.061.0
- 4claude-opus-4-758.8
- 5minimax-m2.756.3
- 6step-3.5-flash21.9
- 7gemini-3.1-pro8.3
- 8deepseek-v4-pro1.4
- 9GLM-5.1-3.5
- 10Kimi K2.6-24.2
| 1 | Qwen3.6-Plus | Qwen | 85.3 | blinded x open-research, 2024-01-01 to 2026-04-10; alpha rank 2; CSI300 buy-and-hold +36.92 | Paper Table 4 Reported by the authors, 27 May 2026 |
|---|---|---|---|---|---|
| 2 | GPT 5.5 | OpenAI | 61.3 | blinded x open-research; alpha rank 6 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 3 | doubao-seed-2.0 | ByteDance | 61.0 | blinded x open-research; alpha rank 3 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 4 | claude-opus-4-7 | Anthropic | 58.8 | blinded x open-research; alpha rank 1 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 5 | minimax-m2.7 | MiniMax | 56.3 | blinded x open-research; alpha rank 4 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 6 | step-3.5-flash | StepFun | 21.9 | blinded x open-research; alpha rank 5 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 7 | gemini-3.1-pro | 8.3 | blinded x open-research; alpha rank 7 | Paper Table 4 Reported by the authors, 27 May 2026 | |
| 8 | deepseek-v4-pro | DeepSeek | 1.4 | blinded x open-research; alpha rank 8 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 9 | GLM-5.1 | Zhipu AI | -3.5 | blinded x open-research; alpha rank 9 | Paper Table 4 Reported by the authors, 27 May 2026 |
| 10 | Kimi K2.6 | Moonshot AI | -24.2 | blinded x open-research; alpha rank 10 | Paper Table 4 Reported by the authors, 27 May 2026 |