Atlas

Benchmarks

← All benchmarks

Deep Research Bench — Compile Dataset

Search · 2025-05-06

The DRB Compile Dataset category asks an agent to assemble a structured dataset from a description of the desired rows and columns.

Top models (higher is better)

ModelScore
Sonnet 4.60.8
GPT-5.50.7
Opus 4.60.7
Opus 4.80.7
Gemini 3.1 Pro Preview0.7
Opus 4.50.7
GPT-50.7
Gemini 3 Pro Preview0.6
Gemini 3 Flash Preview0.6
o30.6
Sonnet 4.50.6
Sonnet 40.5
Gemini 3.1 Flash-Lite Preview0.5
Haiku 4.50.5
Opus 40.5
Opus 4.10.5
Grok 40.5
Gemini 2.5 Pro0.4
GPT-5.10.4
GPT-5.4 Mini0.4
GPT-5.40.4
GPT-5.20.4
Loading Atlas data…