OdysseyArena - AI Trading: leaderboard
Metric: Profit rate in percent averaged over four runs per task (Avg@4) on OdysseyArena-Lite (30 fixed tasks per environment; the agent must induce hidden transition rules from its own interactions; prompts keep past actions and feedback but not earlier reasoning; open models at temperature 0.6), environment AI Trading: trade a multi-asset portfolio whose prices follow hidden latent-factor loadings, with daily news as hints, over 120 steps; cumulative portfolio return after transaction costs (no short selling, so -100 is the floor and there is no upper limit); rules hidden (the benchmark's inductive setting); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 67.71 | #64 |
| 2 | Gemini 2.5 Pro | 33.02 | #145 |
| 3 | GPT-OSS-120B (High) | 23.27 | #330 (GPT-OSS-120B) |
| 4 | GPT-5 (Medium) | 17.32 | #91 (GPT-5) |
| 5 | Qwen 3 235B A22B 2507 Instruct | 11.26 | #291 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 8.62 | #198 (DeepSeek V3.2) |
| 7 | Grok 4 Fast | 5.7 | #242 |
| 8 | Qwen 3 30B A3B 2507 Instruct | 4.76 | #464 |
| 9 | GPT-OSS-120B (Medium) | 3.21 | #330 (GPT-OSS-120B) |
| 10 | GLM-4 32B (0414) | 3.14 | #517 |
| 11 | GPT-OSS-120B (Low) | 2.02 | #330 (GPT-OSS-120B) |
| 12 | Qwen 3 4B 2507 Instruct | 1.67 | #745 |
| 13 | Llama 3.3 70B Instruct | 0.77 | #520 |
| 14 | Llama 3.1 8B Instruct | 0.55 | #1018 |
| 15 | GLM-4 9B Chat | -0.18 | #904 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=odysseyarena-ai-trading · How It Works · Data refreshed daily, snapshot 2026-10-11.