SecRespond - Detection: leaderboard

SecRespond (Alibaba and HKUST, 2026) evaluates LLM agents in the OpenCode harness on post-compromise incident response across 10 compromised cloud-host cyber ranges, investigating a forensic disk snapshot with host security alerts and scans. The score is the detection checkpoint score (discovery, supporting evidence and root-cause attribution) as a percent of the maximum, averaged over ranges and three LLM judges; higher is better.

Metric: Detection CHK-score (%): the detection (discovery, evidence and root-cause attribution, 0-3 per checkpoint) score achieved over each range checkpoints (averaged over Claude Opus 4.7, Gemini 3.1 Pro and GPT-5.4 Pro as judges), averaged over the 10 SecRespond cyber ranges; agents in the OpenCode harness with reasoning enabled investigate a forensic disk snapshot of a compromised host and write forensic reports and a remediation plan; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked.

Top models

#ModelScoreOverall rank
1Claude Opus 4.6 (Thinking)78.2#102 (Claude Opus 4.6)
2Qwen 3.7 Max (Thinking)78#145 (Qwen 3.7 Max)
3GLM-5.176.3#179
4Claude Sonnet 4.6 (Thinking)74.7#161 (Claude Sonnet 4.6)
5GPT-5.4 Pro (xHigh)72.5#85 (GPT-5.4 Pro)
6GPT-5.570.7#49
7Claude Sonnet 4.5 (Thinking)69.7#250 (Claude Sonnet 4.5)
8Claude Opus 4.5 (Thinking)69.5#156 (Claude Opus 4.5)
9Qwen 3.6 Plus (Thinking)68.4#222 (Qwen 3.6 Plus)
10GPT-5.2 Pro66.2#124
11Kimi K2.6 (Thinking)65.8#178 (Kimi K2.6)
12GLM-564.9#260
13Qwen 3.5 Plus (Thinking)63.5#232 (Qwen 3.5 Plus)
14Kimi K2.5 (Thinking)62.1#256 (Kimi K2.5)
15GPT-5.460.7#123

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=secrespond-detection · How It Works · Data refreshed daily, snapshot 2026-11-13.