DSGym - DAEval-Verified: leaderboard
Short data-analysis questions from DAEval, cleaned by DSGym (missing ground truths, misaligned answers and rounding rules fixed) and filtered for questions solvable without the data, run by DSGym's default CodeAct agent (Nie et al., 2026) in an isolated Jupyter container at temperature 0 with all tools disabled. Exact-match accuracy with a small numeric tolerance, in percent; higher is better.
Metric: Exact-Match Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2 0905 | 92.82 | #555 |
| 2 | GPT-4o (2024-08-06) | 92.26 | #731 |
| 3 | Claude Sonnet 4.5 | 91.71 | #244 |
| 4 | Claude Sonnet 4 | 90.91 | #397 |
| 5 | GPT-5 (Medium) | 89.5 | #163 (GPT-5) |
| 6 | GPT-5.1 (High) | 89.5 | #225 (GPT-5.1) |
| 7 | GPT-5.1 (Non-reasoning) | 87.85 | #225 (GPT-5.1) |
| 8 | GPT-OSS-120B | 84.53 | #699 |
| 9 | DeepSeek V3.1 | 82.32 | #500 |
| 10 | Qwen 3 4B Instruct | 64.47 | #1660 |
| 11 | Qwen 2.5 7B Instruct | 50.56 | #2583 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=dsgym-daeval-verified · How It Works · Data refreshed daily, snapshot 2026-11-12.