RHELM (RAG Top-20): leaderboard
Metric: Accuracy (%; question-weighted over 1,305 questions in seven categories, LLM judge, dialogue history plus external sources, dense top-20 retrieval). Source: arxiv.org. Saturation forecast: Around March 2028. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 38.1 |
| 2 | Gemini 2.5 Pro | 35.5 |
| 3 | GPT-4.1 | 29.5 |
| 4 | GPT-4.1 Mini | 25.1 |
Interactive version: theaggregate.ai/benchmark?slug=rhelm-rag-top-20 · How It Works · Data refreshed daily, snapshot 2026-09-25.