RAI-Bench - RAG Robustness (LC Factuality): leaderboard
GovTech Singapore's RAI-Bench asks 331 questions on four Singapore government policy documents, each with its own answer left out of the supplied context. This board is the long in-context setting, where the rest of the knowledge base sits in the prompt. Replies that do not abstain are graded by an LLM judge on a three-tier factuality rubric, and the number is the share judged factual, so higher is better.
Metric: Rate (%). Source: huggingface.co. Saturation forecast: Not forecast. 69 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4 | 56 | #199 |
| 2 | GPT-5.2 | 53 | #110 |
| 3 | Grok 4 | 52 | #181 |
| 4 | O3 | 51 | #124 |
| 5 | Claude Sonnet 4.5 | 50 | #142 |
| 6 | DeepSeek V4 Pro | 50 | #102 |
| 7 | Claude Opus 5 | 49 | #12 |
| 8 | GPT-5 Mini | 48 | #186 |
| 9 | Grok 4.1 Fast | 47 | #217 |
| 10 | GPT-5 | 46 | #98 |
| 11 | Grok 4.1 Fast (Reasoning) | 46 | #217 (Grok 4.1 Fast) |
| 12 | Grok 4.3 | 46 | #188 |
| 13 | Claude Haiku 4.5 | 45 | #280 |
| 14 | O4 Mini | 45 | #180 |
| 15 | Claude Opus 4.6 | 44 | #63 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=rai-bench-rag-robustness-lc-factuality · How It Works · Data refreshed daily, snapshot 2026-10-20.