RHELM (Long Context) - Fact: leaderboard
RHELM category (Zhang et al., Microsoft and Renmin University of China, 2026) that asks for specific facts about the persona scattered through a simulated year of conversation, including multi-hop look-ups, entity disambiguation, attributes at a past state and exclusion constraints. The model reads the whole history, dialogue and external documents in chronological order, in a 1M-token context; accuracy in percent on 207 questions judged 0 or 1 by an LLM, higher is better.
Metric: Accuracy (%; 207 fact questions, LLM judge, dialogue history plus external sources, full history in a 1M-token context). Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 49.3 | #775 |
| 2 | Gemini 2.5 Flash Lite | 31.7 | #958 |
| 3 | Qwen2.5-14B-Instruct-1M | 16.9 | #1417 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rhelm-long-context-fact · How It Works · Data refreshed daily, snapshot 2026-11-12.