RPCBench - Error Group - Underspecified Premise - Evidence Distortion: leaderboard
Metric: Evidence Distortion Rate (%). Source: github.com. Saturation forecast: Around 2030. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.1 8B Instruct | 27.77 |
| 2 | Llama 3.1 70B Instruct | 29.23 |
| 3 | Gemini 3.1 Pro (Preview) | 33.46 |
| 4 | GPT-5.5 | 40.38 |
| 5 | Claude Sonnet 4.6 | 43.08 |
| 6 | Qwen 3.5 35B A3B | 47.15 |
| 7 | Qwen 3.5 122B A10B | 48.31 |
| 8 | DeepSeek V4 Flash | 48.69 |
| 9 | DeepSeek V4 Pro | 48.92 |
| 10 | Qwen 3.5 397B A17B | 49.38 |
| 11 | Qwen 3.5 Plus | 51.08 |
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-error-group-underspecified-premise-evidence-distortion · How It Works · Data refreshed daily, snapshot 2026-09-24.