LITMUS - Chain of Tasks: leaderboard

The Chain of Tasks category of LITMUS (2026), a behavioral jailbreak benchmark for LLM agents in a real Ubuntu system: 31 seed instructions whose harm is judged by the physical OS state after an OpenClaw agent acts. The score is the attack success rate in percent, averaged over three runs; lower is better.

Metric: Attack success rate (%): share of the 31 Chain of Tasks seed cases whose dangerous operation physically took effect on the OS, OpenClaw 4.2.0 agent on Ubuntu 24.04 given seed behavior-jailbreak instructions, physical OS state checked by a GPT-4o verifier and analyzer, mean of three runs per case; lower is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.621.72#173
2GPT-5.3 Codex40.86#124
3Gemini 3.1 Pro (Preview)48.39#99
4Qwen 3.6 Plus56.99#238
5DeepSeek V4 Pro64.52#196
6DeepSeek V3.270.79#419

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=litmus-chain-of-tasks · How It Works · Data refreshed daily, snapshot 2026-11-21.