RHELM (RAG Top-20) - Temporal: leaderboard
RHELM category (Zhang et al., Microsoft and Renmin University of China, 2026) that asks when events happened or in what order, identified through indirect markers or synthesized across long periods. The model answers from the top 20 retrieved chunks of dialogue history and external documents; accuracy in percent on 185 questions judged 0 or 1 by an LLM, higher is better.
Metric: Accuracy (%; 185 temporal questions, LLM judge, dialogue history plus external sources, dense top-20 retrieval). Source: arxiv.org. Saturation forecast: Around October 2028. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 31.9 | #262 |
| 2 | Claude Opus 4.5 | 30.8 | #152 |
| 3 | GPT-4.1 | 29.2 | #492 |
| 4 | GPT-4.1 Mini | 28.7 | #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-temporal · How It Works · Data refreshed daily, snapshot 2026-11-12.