Syslog Severity Classification - RAG: leaderboard
Small and small reasoning language models label real Linux journalctl log entries with their syslog severity level (0-7) with the five most similar labelled training logs retrieved by FAISS over nomic-embed-text embeddings; malformed or verbose answers count as wrong. Accuracy in percent on 9,355 held-out logs, higher is better; Masri et al. (George Mason University), 2026.
Metric: Accuracy on 9,355 held-out journalctl logs, exact syslog severity level (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 4B | 95.64 | #2297 |
| 2 | Qwen 3 0.6B | 88.12 | #8392 |
| 3 | Gemma 3 1B | 85.28 | #9004 |
| 4 | Gemma 3 4B | 81.84 | #5104 |
| 5 | Qwen 3 1.7B | 28.96 | #5551 |
| 6 | DeepSeek R1 Distill Qwen 1.5B | 3.17 | #9257 |
| 7 | Phi-4-mini (Reasoning) | 0 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=syslog-severity-classification-rag · How It Works · Data refreshed daily, snapshot 2026-11-12.