# StockBench

Whether LLM agents can trade a basket of large US stocks profitably over several months. Each day the agent gets prices, fundamentals and news and chooses buy, sell or hold.

## Facts

- Maker: Tsinghua University; Beijing University of Posts and Telecommunications
- Year: 2025
- Use case: Financial trading and portfolios
- Size: 20 DJIA stocks, 82 trading days (2025-03-03 to 2025-06-30), 13 models plus a passive baseline, 3 seeds each
- Public split: Code on GitHub; results on the project page
- Grading: Final return, max drawdown and Sortino ratio, combined into a composite rank by z-score aggregation
- Licence: Apache-2.0 (code); paper says academic, non-commercial research use only
- Submission: maker-only
- Products allowed: Not stated
- Latest version: arXiv v2 (2026-03-02)
- What it does not cover: Markets beyond 20 US large caps, and periods after June 2025
- Last checked: 1 Nov 2026
- Links: paper https://arxiv.org/abs/2510.02209, site https://stockbench.github.io/, code https://github.com/ChenYXxxx/stockbench, leaderboard https://stockbench.github.io/#leaderboard
- Directory page: https://benchmarks.finance/benchmarks/stockbench/
- Leaderboard: https://benchmarks.finance/leaderboards/stockbench/

## Results

### StockBench: final return % (higher is better, 11 models)

| Rank | Model | Provider | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Qwen3-235B-Think | Qwen | 2.5 | final return % | composite rank 4 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 2 | Qwen3-235B-Ins | Qwen | 2.4 | final return % | composite rank 2 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 3 | GLM-4.5 | Zhipu AI | 2.3 | final return % | composite rank 3 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 4 | Claude-4-Sonnet | Anthropic | 2.2 | final return % | composite rank 7 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 5 | Qwen3-30B-Think | Qwen | 2.1 | final return % | composite rank 6 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 6 | Kimi K2 | Moonshot AI | 1.9 | final return % | composite rank 1 (z-score of return, drawdown, Sortino); avg of 3 runs | Reported by the authors | Paper Table 2 (same as project leaderboard) | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 6 | OpenAI-O3 | OpenAI | 1.9 | final return % | composite rank 5 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 8 | DeepSeek-V3.1 | DeepSeek | 1.1 | final return % | composite rank 8 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 9 | GPT-5 | OpenAI | 0.300 | final return % | composite rank 9 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 10 | DeepSeek-V3 | DeepSeek | 0.200 | final return % | composite rank 11 (passive baseline is rank 12 at +0.4%) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |
| 10 | Qwen3-Coder | Qwen | 0.200 | final return % | composite rank 10 | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2510.02209v2 | 2 Mar 2026 (checked 1 Nov 2026) |


Source: https://benchmarks.finance/benchmarks/stockbench/

Created and maintained by Moneda.