Deep Research Bench — Derive Number
Search · 2025-05-06
The DRB Derive Number category asks an agent to derive a requested quantity from information available in the frozen web corpus.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.6 | 0.5 |
| Gemini 3 Flash Preview | 0.4 |
| Opus 4.6 | 0.4 |
| Gemini 3.1 Pro Preview | 0.4 |
| Opus 4.5 | 0.4 |
| GPT-5.5 | 0.4 |
| Opus 4.8 | 0.4 |
| Gemini 3 Pro Preview | 0.3 |
| GPT-5 | 0.3 |
| Opus 4.1 | 0.3 |
| Sonnet 4.5 | 0.3 |
| Gemini 3.1 Flash-Lite Preview | 0.3 |
| Haiku 4.5 | 0.3 |
| Opus 4 | 0.3 |
| Sonnet 4 | 0.3 |
| GPT-5.2 | 0.2 |
| Gemini 2.5 Pro | 0.2 |
| GPT-5.4 | 0.2 |
| Grok 4 | 0.2 |
| o3 | 0.2 |
| GPT-5.1 | 0.2 |
| GPT-5.4 Mini | 0.2 |