CreditCardQA: leaderboard
Financial numerical reasoning from credit-card agreements. Reports Chain-of-Thought accuracy with a strict ±0.2% numerical tolerance; model sizes remain distinct.
Metric: Chain-of-Thought Accuracy (%, ±0.2% tolerance). Source: arxiv.org. Saturation forecast: Around January 2028. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 70.5 | #80 |
| 2 | GPT-5 | 70.4 | #98 |
| 3 | GPT-OSS-120B | 70 | #341 |
| 4 | GPT-5 Mini | 70 | #186 |
| 5 | GPT-OSS-20B | 66.8 | #518 |
| 6 | GLM-4.6 | 65 | #252 |
| 7 | Kimi K2 (Thinking) | 64.1 | #243 (Kimi K2) |
| 8 | QwQ-32B | 63.6 | #421 |
| 9 | DeepSeek R1 | 63.3 | #251 |
| 10 | Llama 3.3 70B | 62 | #579 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=creditcardqa · How It Works · Data refreshed daily, snapshot 2026-10-20.