DABstep: leaderboard
DABstep validated agent submissions: hard-level accuracy on multi-step data analysis questions requiring CSV analysis and document cross-references. Rows identify submitted agent systems; reported model families do not establish a shared checkpoint or harness.
Metric: Hard Level Accuracy (%). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DABstep NVIDIA KGMON (NeMo Agent Toolkit) Data Explorer | 89.95 | |
| 2 | DABstep DataPilot | 87.57 | |
| 3 | DABstep gg-agent-gpt5-1104-1 | 62.96 | |
| 4 | DABstep CambioML energent.ai DS Agent | 57.67 | |
| 5 | DABstep DS-STAR | 45.24 | |
| 6 | DABstep Amity DA Agent v0.1 | 41.01 | |
| 7 | DABstep AgenticData | 40.48 | |
| 8 | DABstep Mphasis-I2I-Agents | 28.04 | |
| 9 | DABstep Claude 4 Sonnet ReACT Baseline | 19.84 | |
| 10 | DABstep Open Data Scientist | 16.4 |
Interactive version: theaggregate.ai/benchmark?slug=dabstep · How It Works · Data refreshed daily, snapshot 2026-10-20.