RAI-Bench - RAG Robustness (HY Abstention): leaderboard
Metric: Rate (%). Source: huggingface.co. 70 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Flash Lite | 93 |
| 2 | DeepSeek V4 Pro | 91 |
| 3 | DeepSeek V4 Flash | 91 |
| 4 | Grok 4.3 | 91 |
| 5 | Claude Sonnet 4.5 | 90 |
| 6 | GPT-5.4 Mini | 90 |
| 7 | Gemini 3.5 Flash | 90 |
| 8 | Gemini 3.6 Flash | 90 |
| 9 | Claude Sonnet 4 | 89 |
| 10 | GLM-5 | 89 |
| 11 | Gemini 3.5 Flash Lite | 89 |
| 12 | Grok 4.1 Fast (Reasoning) | 88 |
| 13 | Grok 4 | 88 |
| 14 | GLM-4.6 | 88 |
| 15 | Claude Opus 5 | 88 |
Interactive version: theaggregate.ai/benchmark?slug=rai-bench-rag-robustness-hy-abstention · How It Works · Data refreshed daily, snapshot 2026-09-05.