# DeepSeek-V3.1 vs GPT-5

Every finance benchmark where both have a score (DeepSeek and OpenAI). Data refreshed 2 Nov 2026. Scores from different benchmarks use different scales and are not added up.

| Benchmark | Metric | DeepSeek-V3.1 | DeepSeek-V3.1 source | GPT-5 | GPT-5 source | Ahead |
| --- | --- | --- | --- | --- | --- | --- |
| AI-Trader | cumulative return % (higher is better) | 8.4 | Paper Table 1, 1 Dec 2025, https://arxiv.org/html/2512.10971v1 | 1.6 | Paper Table 1, 1 Dec 2025, https://arxiv.org/html/2512.10971v1 | DeepSeek-V3.1 |
| CAIA (Crypto AI Agent Benchmark) | accuracy % (majority vote, 5 runs) (higher is better) | 49.2 | Paper Tables 2-3, 17 Jan 2026, https://arxiv.org/html/2510.00332v2 | 67.4 | Paper Tables 2-3, 17 Jan 2026, https://arxiv.org/html/2510.00332v2 | GPT-5 |
| Reasoning Models Ace the CFA Exams (CFA mock-exam evaluation) | Level I accuracy % (higher is better) | 90.9 | Paper Table 5, 9 Dec 2025, https://arxiv.org/html/2512.08270v1 | 96.1 | Paper Table 5, 9 Dec 2025, https://arxiv.org/html/2512.08270v1 | GPT-5 |
| Reasoning Models Ace the CFA Exams (CFA mock-exam evaluation) | Level III constructed-response score % (higher is better) | 70.8 | Paper Table 5, 9 Dec 2025, https://arxiv.org/html/2512.08270v1 | 71.8 | Paper Table 5, 9 Dec 2025, https://arxiv.org/html/2512.08270v1 | GPT-5 |
| StockBench | final return % (higher is better) | 1.1 | Paper Table 2, 2 Mar 2026, https://arxiv.org/html/2510.02209v2 | 0.300 | Paper Table 2, 2 Mar 2026, https://arxiv.org/html/2510.02209v2 | DeepSeek-V3.1 |

Source: https://benchmarks.finance/compare/deepseek-v3-1-vs-gpt-5/

Created and maintained by Moneda.