SAGE (Literature Search) - Open-Ended (Computer Science): leaderboard
Metric: Weighted recall (%, 0-100) of the ground-truth papers found in the answer text or citations, the two seed papers weighted 2 and their shared references 1; the 150 computer science open-ended related-work questions of SAGE; deep research agents with their native web search: GPT-5 series through the API at medium reasoning effort with web search enabled, Gemini 2.5 with dynamic thinking and web search, the open DR Tulu agent served with vLLM; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 35.1 | #91 |
| 2 | Gemini 2.5 Flash | 28.2 | #237 |
| 3 | GPT-5 Mini | 27.4 | #176 |
| 4 | GPT-5 Nano | 25.9 | #415 |
| 5 | Gemini 2.5 Pro | 20 | #145 |
| 6 | dr-tulu | 18 | #188 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=sage-literature-search-open-ended-computer-science · How It Works · Data refreshed daily, snapshot 2026-10-11.