CiteTrace - Suitability Failure Rate: leaderboard
Whether search-augmented LLM APIs cite source types that suit the domain of the question, on CiteTrace (Seo et al., 2026, Verified Misguidance): 11,200 real Stack Exchange questions from 28 communities with every cited page crawled. Percent of evaluable citations whose source type (official, research, company, blog and others) is rated 1-2 for the domain on an expert-validated suitability matrix, classified by gpt-4o-mini; lower is better. The paper finds that the provider search backend, more than the model, drives this rate.
Metric: Suitability failure rate (%; share of evaluable citations whose source type is rated 1-2 for the question's domain on the expert-validated 10x6 source-suitability matrix; gpt-4o-mini classifier; 11,200 Stack Exchange questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 7.98 | #163 |
| 2 | GPT-5 Mini | 14.25 | #349 |
| 3 | Grok 4.1 Fast (Reasoning) | 24.56 | #426 (Grok 4.1 Fast) |
| 4 | Grok 4.1 Fast (Non-reasoning) | 24.79 | #426 (Grok 4.1 Fast) |
| 5 | Sonar | 29.99 | #1677 |
| 6 | Gemini 3.1 Pro (Preview) | 30.07 | #91 |
| 7 | Claude Haiku 4.5 | 30.07 | #568 |
| 8 | Claude Sonnet 4.6 | 31.65 | #158 |
| 9 | Gemini 3 Flash | 33.46 | #159 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=citetrace-suitability-failure-rate · How It Works · Data refreshed daily, snapshot 2026-11-12.