RAI-Bench - Refusal Rate (JD): leaderboard
GovTech Singapore's RAI-Bench runs adversarial prompts from RabakBench in four Singapore deployment scenarios, fifty per scenario. This board is the job description writer scenario, where the requested criteria breach Singapore's Tripartite Guidelines on Fair Employment Practices. An LLM judge labels each reply Reject, Reframe or Engage, and the number is the share labelled Reject or Reframe, so higher is better.
Metric: Rate (%). Source: huggingface.co. 78 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 100 | #245 |
| 2 | Claude Opus 4.6 | 100 | #63 |
| 3 | Gemma 4 31B | 100 | #200 |
| 4 | Claude Opus 4 | 100 | #163 |
| 5 | Claude Sonnet 4.6 | 98 | #90 |
| 6 | Claude Haiku 4.5 | 98 | #280 |
| 7 | Grok 3 | 98 | #304 |
| 8 | Claude Sonnet 4.5 | 96 | #142 |
| 9 | Claude Sonnet 4 | 96 | #199 |
| 10 | Gemini 2.5 Flash Lite | 96 | #422 |
| 11 | Gemini 3.1 Flash Lite | 96 | #191 |
| 12 | GLM-4.7 | 96 | #192 |
| 13 | Gemini 3.5 Flash Lite | 96 | #156 |
| 14 | Claude Opus 4.5 | 94 | #83 |
| 15 | Grok 4.1 Fast | 94 | #217 |
Interactive version: theaggregate.ai/benchmark?slug=rai-bench-refusal-rate-jd · How It Works · Data refreshed daily, snapshot 2026-10-20.