RPCBench - Evidence Distortion: leaderboard
Metric: Evidence Distortion Rate (%). Source: arxiv.org. Saturation forecast: Around October 2027. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 16.4 |
| 2 | Gemini 3.1 Pro (Preview) | 17.7 |
| 3 | Qwen 3.5 397B A17B | 21.1 |
| 4 | Qwen 3.5 122B A10B | 21.3 |
| 5 | Qwen 3.5 Plus | 21.5 |
| 6 | Qwen 3.5 35B A3B | 22.7 |
| 7 | Llama 3.1 8B Instruct | 23.2 |
| 8 | Claude Sonnet 4.6 | 24.2 |
| 9 | Llama 3.1 70B Instruct | 25.8 |
| 10 | DeepSeek V4 Flash | 26.4 |
| 11 | DeepSeek V4 Pro | 26.6 |
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-evidence-distortion · How It Works · Data refreshed daily, snapshot 2026-09-24.