SDABench - Exploratory: leaderboard

Metric: Accuracy (%) of open-ended answers on the held-out SDA-Synth test split (the 400 exploratory analysis instances; zero-shot, JSON reasoning and answer fields, the answer scored by relative-error numeric, exact or structured matching with a GPT-4o fallback judge); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 13 models tracked.

Top models

#ModelScore
1GPT-5.453.75
2Claude Sonnet 4.650.25
3Gemini 3.1 Pro (Preview)48.25
4GPT-5.4 Mini45.25
5DeepSeek V3.241.5
6Qwen 3.5 397B A17B37
7GLM-536.5
8DeepSeek R134.5
9Kimi K2.529.5
10Qwen 3 235B A22B 2507 Instruct27.75
11Llama 3.3 70B Instruct11.75
12Llama 3.1 8B Instruct8.25

Interactive version: theaggregate.ai/benchmark?slug=sdabench-exploratory · How It Works · Data refreshed daily, snapshot 2026-09-29.