Cyber-harness network challenges (Incalmo cyber ranges)
Agents run long-horizon attacks on cyber ranges of 25 to 50 hosts using a specialized offensive harness.
Built with Carnegie Mellon University, as described in the Incalmo paper. The ranges are not public.
- Level
- Research frontier Researchers, on open problems or tasks that take days
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Claude 3.7 Sonnet leads with 66.7%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 66.7% | Feb 2025 | Model card |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
3 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 29 Sep 2025 | Claude Sonnet 4.5 system card | Incalmo cyber ranges, bash shell on Kali without the Incalmo harness, no overall. Asset acquisition on all 4 ranges with any model success (Equifax, 4-Chain, Dumbbell A, 4-Star), none on 5 others. Comprehensiveness 1.00, 0.32, 1.00, 0.48. Reliability 2, 3, 4, 3 of 5 trials (Figures 5.3.2.A to 5.3.2.C) | Cyber risk | |
| 22 May 2025 | Claude Opus 4 system card | Cyber-harness network challenges (Incalmo, three ranges), no number printed. Text says generally higher than Claude Sonnet 3.7 on all three ranges (7.4.6) | Cyber risk | |
| 24 Feb 2025 | Claude 3.7 Sonnet system card | 66.7 | Cyber-harness network challenges (three cyber ranges), 2/3 solved, pass@5 (1/1 easy, 1/2 medium) (7.3.2, Figure 24) | Cyber risk |
More in Cybersecurity
- ExploitGym In use 869 real-vulnerability exploit-development tasks (userspace, V8, Linux kernel). Success means capturing a flag through the intended vulnerability. GPT-6 Astra 42.4%
- ExploitBench In use 41 recent V8 JavaScript-engine N-day vulnerabilities. Progress measured by 16 grader-verified capability flags across five tiers. GPT-6 Astra 100%
- CWE-bench In use Agents audit a repository and write a patch for a security flaw, without being told where the bug is. Covers six programming languages.
Gemini 4 Carbon 71% - Real-world vulnerability discovery In use Agents search popular open-source codebases to rediscover about 1,200 recent confirmed vulnerabilities across many languages. Reports share discovered. Gemini 4 Argon 86%
- Artificial Analysis Cyber Index In use Agents do enterprise cyber defense tasks from three benchmarks, CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA, combined into one success rate.
Mistral Large 4 (Le Chonk, open weights) 51% - Big Sleep Evaluation (Google) In use Agents search large real codebases such as Chrome and Safari for critical, hard to find vulnerabilities. Scored by pass@1 success rate, run without safety guardrails. Gemini 3.5 Flash Cyber 72%