RPCBench - Error Group - Inconsistent 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 397B A17B | 0.7 |
| 2 | Qwen 3.5 Plus | 0.7 |
| 3 | Qwen 3.5 122B A10B | 0.86 |
| 4 | Qwen 3.5 35B A3B | 1.29 |
| 5 | GPT-5.5 | 2.84 |
| 6 | Gemini 3.1 Pro (Preview) | 3.43 |
| 7 | DeepSeek V4 Pro | 3.86 |
| 8 | DeepSeek V4 Flash | 4.82 |
| 9 | Claude Sonnet 4.6 | 5.14 |
| 10 | Llama 3.1 70B Instruct | 46.44 |
| 11 | Llama 3.1 8B Instruct | 55.01 |
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-error-group-inconsistent-premise-fact-fabrication · How It Works · Data refreshed daily, snapshot 2026-09-24.