Benchmark
Sayf RedSage: cybersecurity_skills accuracy
Safety benchmarks examine risks in model behaviour, including how models respond to unsafe requests. Higher scores are better.
- Models measured
- 6
- Scores recorded
- 6
- Sources
- 1
- IQ weight
- 0.77
Score distribution
- 0.916
- Mean
- 0.916
- Median
- 0.015
- Std dev
- 0.902
- 10th percentile
- 0.936
- 90th percentile
- 0.894
- Min
This benchmark has a weight of 0.77 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 | Claude Sonnet 4.6redsage_skills· Sayf cybersecurity evaluations data docs | 0.936 |
| 2 | GPT 5.4redsage_skills· Sayf cybersecurity evaluations data docs | 0.932 |
| 3 | Qwen3.6 35B A3Bredsage_skills· Sayf cybersecurity evaluations data docs | 0.916 |
| 4 | Gemma 4 31B ITredsage_skills· Sayf cybersecurity evaluations data docs | 0.914 |
| 5 | GPT OSS 20bredsage_skills· Sayf cybersecurity evaluations data docs | 0.902 |
| 6 | Llama 3.3 70B Instructredsage_skills· Sayf cybersecurity evaluations data docs | 0.894 |
Data credits: Sayf-eval. Copyright (c) 2026 Qatar Computing Research Institute (QCRI). MIT License. (licence)