RPCBench - Error Group - Boundary-Violating Premise - Conditional Localization: leaderboard
RPCBench tests proactive premise critique in recommendation requests. This series measures conditional localization on the Boundary-Violating Premise subset (436 questions), using the same eleven-model, three-judge experiment as the headline paper. Published CLA fractions from the authors' per-model CSV are scaled once to Localization Accuracy (%). Higher is better; domains and nested premise-error groups are related slices of the same experiment.
Metric: Localization Accuracy (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 97.4 | #232 |
| 2 | Qwen 3.5 397B A17B | 96.82 | #245 |
| 3 | GPT-5.5 | 95.5 | #48 |
| 4 | Claude Sonnet 4.6 | 95.42 | #159 |
| 5 | Gemini 3.1 Pro (Preview) | 94.31 | #93 |
| 6 | Qwen 3.5 122B A10B | 93.75 | #335 |
| 7 | Qwen 3.5 35B A3B | 93.03 | #492 |
| 8 | DeepSeek V4 Flash | 83.42 | #259 |
| 9 | DeepSeek V4 Pro | 83.33 | #188 |
| 10 | Llama 3.1 70B Instruct | 81.29 | #1629 |
| 11 | Llama 3.1 8B Instruct | 59.45 | #4318 |
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-boundary-violating-premise-conditional-localization · How It Works · Data refreshed daily, snapshot 2026-11-08.