Benchmark
Sayf CTI-Bench VSP (CVSS) accuracy
Higher scores are better.
Benchmark sourceScores from Sayf cybersecurity evaluations data docs.
- Models measured
- 6
- Scores recorded
- 6
- Sources
- 1
- IQ weight
- 0.70
Score distribution
- 0.344
- Mean
- 0.423
- Median
- 0.107
- Std dev
- 0.227
- 10th percentile
- 0.486
- 90th percentile
- 0.221
- Min
This benchmark has a weight of 0.70 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 31B ITvsp· Sayf cybersecurity evaluations data docs | 0.486 |
| 2 | GPT 5.4vsp· Sayf cybersecurity evaluations data docs | 0.436 |
| 3 | Claude Sonnet 4.6vsp· Sayf cybersecurity evaluations data docs | 0.423 |
| 4 | Llama 3.3 70B Instructvsp· Sayf cybersecurity evaluations data docs | 0.273 |
| 5 | Qwen3.6 35B A3Bvsp· Sayf cybersecurity evaluations data docs | 0.227 |
| 6 | GPT OSS 20bvsp· Sayf cybersecurity evaluations data docs | 0.221 |
Data credits: Sayf-eval. Copyright (c) 2026 Qatar Computing Research Institute (QCRI). MIT License. (licence)