RPCBench - Error Group - Underspecified Premise - Fact Fabrication: leaderboard
Metric: Fact Fabrication Rate (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 35B A3B | 2.77 |
| 2 | Qwen 3.5 Plus | 2.92 |
| 3 | Qwen 3.5 397B A17B | 3 |
| 4 | Qwen 3.5 122B A10B | 3.15 |
| 5 | GPT-5.5 | 7.77 |
| 6 | DeepSeek V4 Flash | 9.77 |
| 7 | DeepSeek V4 Pro | 11.31 |
| 8 | Gemini 3.1 Pro (Preview) | 17.15 |
| 9 | Claude Sonnet 4.6 | 21.38 |
| 10 | Llama 3.1 70B Instruct | 40.77 |
| 11 | Llama 3.1 8B Instruct | 46.85 |
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-error-group-underspecified-premise-fact-fabrication · How It Works · Data refreshed daily, snapshot 2026-09-24.