# AI-Trader leaderboard

Fully automated live trading by LLM agents in US stocks, China A-shares and crypto. Agents start with minimal context and must search for and verify market information themselves.

## AI-Trader: cumulative return % (higher is better, 6 models)

| Rank | Model | Provider | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | MiniMax-M2 | MiniMax | 9.6 | cumulative return % | US market (Nasdaq-100, hourly), 2025-10-01 to 2025-11-07; QQQ baseline 1.87 | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |
| 2 | DeepSeek-V3.1 | DeepSeek | 8.4 | cumulative return % | US market, hourly | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |
| 3 | Claude-3.7-Sonnet | Anthropic | 3.1 | cumulative return % | US market, hourly | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |
| 4 | GPT-5 | OpenAI | 1.6 | cumulative return % | US market, hourly | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |
| 5 | Qwen3-Max | Qwen | 0.390 | cumulative return % | US market, hourly | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |
| 6 | Gemini 2.5 Flash | Google | -0.060 | cumulative return % | US market, hourly | Reported by the authors | Paper Table 1 | https://arxiv.org/html/2512.10971v1 | 1 Dec 2025 (checked 1 Nov 2026) |

Scores are never combined across benchmarks. Each row names its metric, source and date.

## About this benchmark

- Maker: The University of Hong Kong (HKUDS)
- Year: 2025
- Use case: Financial trading and portfolios
- Size: 3 markets (Nasdaq-100 hourly, SSE-50 daily, 10 crypto pairs daily), 6 LLMs in the paper
- Public split: Code and evaluation data on GitHub; live platform at ai4trade.ai
- Grading: Cumulative return, Sortino ratio, max drawdown vs market index baseline
- Licence: MIT (code, per README badge)
- Submission: open
- Products allowed: Yes
- Latest version: arXiv v1 (2025-12-01); repo updated 2026-06-11
- What it does not cover: Long horizons (paper windows are 2-5 weeks); the repo now presents itself as a trading platform more than a fixed benchmark
- Last checked: 1 Nov 2026
- Links: paper https://arxiv.org/abs/2512.10971, site https://ai4trade.ai, code https://github.com/HKUDS/AI-Trader
- Directory page: https://benchmarks.finance/benchmarks/ai-trader/
- Leaderboard: https://benchmarks.finance/leaderboards/ai-trader/

Source: https://benchmarks.finance/leaderboards/ai-trader/

Created and maintained by Moneda.