Benchmark
Sayf RedSage: cybersecurity_knowledge_generals 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
- 0.91
Score distribution
- 0.871
- Mean
- 0.871
- Median
- 0.029
- Std dev
- 0.858
- 10th percentile
- 0.907
- 90th percentile
- 0.821
- Min
This benchmark has a weight of 0.91 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_generals· Sayf cybersecurity evaluations data docs | 0.907 |
| 2 | GPT 5.4redsage_generals· Sayf cybersecurity evaluations data docs | 0.902 |
| 3 | Gemma 4 31B ITredsage_generals· Sayf cybersecurity evaluations data docs | 0.871 |
| 4 | Qwen3.6 35B A3Bredsage_generals· Sayf cybersecurity evaluations data docs | 0.867 |
| 5 | Llama 3.3 70B Instructredsage_generals· Sayf cybersecurity evaluations data docs | 0.858 |
| 6 | GPT OSS 20bredsage_generals· Sayf cybersecurity evaluations data docs | 0.821 |
Data credits: Sayf-eval. Copyright (c) 2026 Qatar Computing Research Institute (QCRI). MIT License. (licence)