RPCBench - Domain - Yelp Local - Fact Fabrication: leaderboard
RPCBench tests proactive premise critique in recommendation requests. This series measures fact fabrication on the Yelp Local subset (1,310 questions), using the same eleven-model, three-judge experiment as the headline paper. Published FFR fractions from the authors' per-model CSV are scaled once to Fact Fabrication Rate (%). Lower is better; domains and nested premise-error groups are related slices of the same experiment.
Metric: Fact Fabrication Rate (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 1.15 | #227 |
| 2 | Qwen 3.5 397B A17B | 1.3 | #242 |
| 3 | Qwen 3.5 35B A3B | 1.37 | #480 |
| 4 | Qwen 3.5 122B A10B | 2.21 | #329 |
| 5 | DeepSeek V4 Flash | 4.58 | #254 |
| 6 | GPT-5.5 | 5.11 | #48 |
| 7 | DeepSeek V4 Pro | 5.5 | #181 |
| 8 | Gemini 3.1 Pro (Preview) | 6.72 | #91 |
| 9 | Claude Sonnet 4.6 | 9.69 | #158 |
| 10 | Llama 3.1 70B Instruct | 53.97 | #1541 |
| 11 | Llama 3.1 8B Instruct | 62.37 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rpcbench-domain-yelp-local-fact-fabrication · How It Works · Data refreshed daily, snapshot 2026-11-12.