Atlas

Benchmarks

← All benchmarks

Deep Research Bench — Find Original Source

Search · 2025-05-06

The DRB Find Original Source category asks an agent to trace a claim back to its original published source.

Top models (higher is better)

ModelScore
Opus 4.50.6
GPT-5.50.6
Opus 4.60.5
Opus 4.80.5
Sonnet 4.60.5
Haiku 4.50.4
Opus 40.4
Sonnet 40.4
Gemini 3.1 Pro Preview0.4
Gemini 3 Flash Preview0.4
GPT-50.4
Grok 40.4
Sonnet 4.50.4
Opus 4.10.4
Gemini 3 Pro Preview0.4
GPT-5.10.4
o30.4
Gemini 2.5 Pro0.3
GPT-5.20.3
GPT-5.4 Mini0.3
Gemini 3.1 Flash-Lite Preview0.3
GPT-5.40.3
Loading Atlas data…