RHELM (RAG Top-20) - External Source: leaderboard
RHELM category (Zhang et al., Microsoft and Renmin University of China, 2026) that asks about the persona's documents and emails, including facts in attachments, reasoning and aggregation over tables, navigation by document structure, and counting or locating emails over a period. The model answers from the top 20 retrieved chunks of dialogue history and external documents; accuracy in percent on 249 questions judged 0 or 1 by an LLM, higher is better.
Metric: Accuracy (%; 249 external-source (attachment and email) questions, LLM judge, dialogue history plus external sources, dense top-20 retrieval). Source: arxiv.org. Saturation forecast: Around July 2028. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 33.7 | #152 |
| 2 | GPT-4.1 | 32.5 | #492 |
| 3 | Gemini 2.5 Pro | 31.3 | #262 |
| 4 | GPT-4.1 Mini | 28.5 | #775 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rhelm-rag-top-20-external-source · How It Works · Data refreshed daily, snapshot 2026-11-12.