EnterpriseRAG Bench - Correctness: leaderboard
The "Correctness" subtask of EnterpriseRAG Bench. Onyx's enterprise retrieval-augmented generation benchmark — question answering over realistic company document corpora, scoring answer correctness, completeness, recall and resistance to plausible-but-irrelevant extra documents.
Metric: Score (%). Source: huggingface.co. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Flexor.ai (+ GPT‑6.1 Sol) | 93 | |
| 2 | CDL (Causal Dynamics Lab) | 92.4 | |
| 3 | Mixedbread (+ Opus 5) | 89.8 | |
| 4 | Neurond AI (neurond.com) | 87.4 | |
| 5 | Prism (aurait.ai) | 85.6 | |
| 6 | PERSONA AI (personaai.co.kr) | 84.4 | |
| 7 | Troml | 83.8 | |
| 8 | ZNV AgentCube | 82.4 | |
| 9 | metor.com | 82 | |
| 10 | Skyller AI | 81.6 |
Interactive version: theaggregate.ai/benchmark?slug=enterpriserag-bench-correctness · How It Works · Data refreshed daily, snapshot 2026-10-20.