RPCBench - Error Group - Underspecified Premise - Proactive Detection: leaderboard
RPCBench tests proactive premise critique in recommendation requests. This series measures proactive detection on the Underspecified Premise subset (1,300 questions), using the same eleven-model, three-judge experiment as the headline paper. Published PDR fractions from the authors' per-model CSV are scaled once to Detection Rate (%). Higher is better; domains and nested premise-error groups are related slices of the same experiment.
Metric: Detection Rate (%). Source: github.com. Saturation forecast: Around 2032. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Pro | 15.15 | #188 |
| 2 | Claude Sonnet 4.6 | 11.77 | #159 |
| 3 | Qwen 3.5 35B A3B | 11.77 | #492 |
| 4 | Gemini 3.1 Pro (Preview) | 11.54 | #93 |
| 5 | DeepSeek V4 Flash | 11.08 | #259 |
| 6 | Qwen 3.5 122B A10B | 11.08 | #335 |
| 7 | Qwen 3.5 Plus | 10.77 | #232 |
| 8 | Qwen 3.5 397B A17B | 9.77 | #245 |
| 9 | GPT-5.5 | 8.38 | #48 |
| 10 | Llama 3.1 8B Instruct | 5 | #4318 |
| 11 | Llama 3.1 70B Instruct | 3.85 | #1629 |
No result here: #4 Claude Opus 5.5, #5 GPT-6 Astra, #9 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-error-group-underspecified-premise-proactive-detection · How It Works · Data refreshed daily, snapshot 2026-11-08.