RPCBench - Error Group - Inconsistent Premise - Composite Premise Critique: leaderboard
RPCBench tests proactive premise critique in recommendation requests. This series measures composite premise critique on the Inconsistent Premise subset (1,867 questions), using the same eleven-model, three-judge experiment as the headline paper. Published CPCC fractions from the authors' per-model CSV are scaled once to Composite Score (0-100). Higher is better; domains and nested premise-error groups are related slices of the same experiment.
Metric: Composite Score (0-100). Source: github.com. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Pro | 69.63 | #188 |
| 2 | Claude Sonnet 4.6 | 68.73 | #159 |
| 3 | Qwen 3.5 397B A17B | 68.44 | #245 |
| 4 | Qwen 3.5 Plus | 67.95 | #232 |
| 5 | Qwen 3.5 122B A10B | 67.65 | #335 |
| 6 | Gemini 3.1 Pro (Preview) | 66.78 | #93 |
| 7 | DeepSeek V4 Flash | 62.99 | #259 |
| 8 | Qwen 3.5 35B A3B | 61.9 | #492 |
| 9 | GPT-5.5 | 61.51 | #48 |
| 10 | Llama 3.1 70B Instruct | 14.04 | #1629 |
| 11 | Llama 3.1 8B Instruct | 10.03 | #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-inconsistent-premise-composite-premise-critique · How It Works · Data refreshed daily, snapshot 2026-11-08.