From Knowing to Doing: leaderboard
Evaluating whether large language model (LLM) agents can profit in capital markets is increasingly framed as end-to-end trading: place an agent in a historical market, let it trade, and measure portfolio returns. This setup is vulnerable to two evaluation failures. First, long backtests often overlap with the knowledge cutoffs of frontier LLMs, allowing memorized tickers, dates, prices, and market
Metric: Total ret. (self-reported). Source: benchmarklist.com. Saturation forecast: Not forecast. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 85.29 | #120 |
| 2 | GPT-5.5 | 61.26 | #31 |
| 3 | Claude Opus 4.7 | 58.8 | #49 |
| 4 | MiniMax-M2.7 | 56.31 | #258 |
| 5 | Step 3.5 Flash | 21.86 | #270 |
| 6 | Gemini 3.1 Pro (Preview) | 8.25 | #58 |
| 7 | DeepSeek V4 Pro | 1.39 | #102 |
| 8 | GLM-5.1 | -3.55 | #97 |
| 9 | Kimi K2.6 | -24.23 | #104 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=from-knowing-to-doing · How It Works · Data refreshed daily, snapshot 2026-10-20.