SecRespond - Persistence Mechanism Detection: leaderboard

Metric: Detection CAP-score (%) on the Persistence Mechanism (PER) capability dimension: the detection (discovery, evidence and root-cause attribution, 0-3 per checkpoint) score achieved over all checkpoints mapped to the dimension across the 10 SecRespond cyber ranges (averaged over three LLM judges); agents in the OpenCode harness with reasoning enabled; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 23 models tracked.

Top models

#ModelScore
1Qwen 3.7 Max (Thinking)79
2GPT-5.4 Pro (xHigh)76.4
3Claude Opus 4.6 (Thinking)72.1
4GPT-5.571.3
5GLM-5.168.4
6GPT-5.2 Pro67.6
7Claude Sonnet 4.6 (Thinking)58.7
8Kimi K2.6 (Thinking)57.9
9Qwen 3.6 Plus (Thinking)57.5
10Claude Opus 4.5 (Thinking)56.9
11Claude Sonnet 4.5 (Thinking)56.1
12GLM-554.2
13Kimi K2.5 (Thinking)49.5
14Qwen 3.5 Plus (Thinking)49.5
15Gemini 3.1 Pro (Preview)48.7

Interactive version: theaggregate.ai/benchmark?slug=secrespond-persistence-mechanism-detection · How It Works · Data refreshed daily, snapshot 2026-09-29.