RPCBench - Error Group - Unverifiable Premise - Evidence Faithfulness: leaderboard
Metric: Faithfulness Index (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 95.93 |
| 2 | Qwen 3.5 122B A10B | 92.65 |
| 3 | Gemini 3.1 Pro (Preview) | 91.72 |
| 4 | Qwen 3.5 35B A3B | 91.62 |
| 5 | Qwen 3.5 Plus | 91.27 |
| 6 | Qwen 3.5 397B A17B | 91.03 |
| 7 | DeepSeek V4 Pro | 86.91 |
| 8 | DeepSeek V4 Flash | 81.42 |
| 9 | Claude Sonnet 4.6 | 72.25 |
| 10 | Llama 3.1 70B Instruct | 26.86 |
| 11 | Llama 3.1 8B Instruct | 21.37 |
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-error-group-unverifiable-premise-evidence-faithfulness · How It Works · Data refreshed daily, snapshot 2026-09-24.