Benchmark Almanac

Pwn CTF challenges (Anthropic)

Claude Opus 4.6 71.4% leads since Feb 2026 Only in system cards

Agents exploit insecure software on a remote server for privilege escalation or code execution in capture-the-flag challenges.

In-house Anthropic evaluation. The tasks are not public.

Made by Anthropic Released Feb 2025 Capture-the-flag challenges Tasks kept private Last in a system card Feb 2026 Last model added Feb 2026
Level
Expert Specialists with years of training, often over hours of work
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.
6
Models scored in lab cards
7
System cards report it, from 1 lab

2 models share the top score of 71.4% among models released by Feb 2026

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.

The newest model with a score here came out in Feb 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 71.4 Claude Opus 4.6
  2. 71.4 Claude Sonnet 4.6
  3. 55.6 Claude Opus 4
  4. 42.9 Claude Opus 4.5
  5. 38.5 Claude 3.7 Sonnet
  6. 33.3 Claude Sonnet 4
  1. 1 Claude Opus 4.6 71.4%
  2. 2 Claude Sonnet 4.6 71.4%
  3. 3 Claude Opus 4 55.6%
  4. 4 Claude Opus 4.5 42.9%
  5. 5 Claude 3.7 Sonnet 38.5%
  6. 6 Claude Sonnet 4 33.3%
#ModelScoreSettingDateSource
1 Claude Opus 4.6 71.4% Feb 2026 Model card
2 Claude Sonnet 4.6 71.4% Feb 2026 Model card
3 Claude Opus 4 55.6% May 2025 Model card
4 Claude Opus 4.5 42.9% Nov 2025 Model card
5 Claude 3.7 Sonnet 38.5% Feb 2025 Model card
6 Claude Sonnet 4 33.3% May 2025 Model card

3 models have held number one since its release in Feb 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Lab claims Not tested since Feb 2026
Claude Opus 4.6 71.4%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Opus 4.6 Claude Opus 4 Claude 3.7 Sonnet 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Claude Opus 4.6 Claude Opus 4

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.

7 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.

DateDocumentScoreAs the document writes itSection
17 Feb 2026 Claude Sonnet 4.6 system card 71.4 Pwn CTF, 5/7 challenges solved (Figure 6.4.4.A) Cyber risk
5 Feb 2026 Claude Opus 4.6 system card 71.4 Pwn CTF challenges, 5/7 solved, pass@30 (Figure 8.4.4.A); details text lists 9 challenges Cyber risk
24 Nov 2025 Claude Opus 4.5 system card 42.9 Pwn CTF challenges, 3/7 solved, pass@30 (Figure 7.4.4.A) Cyber risk
29 Sep 2025 Claude Sonnet 4.5 system card Irregular private CTF challenges, Pwn category, mean pass@1, chart without printed values (5.3.1, Figure 5.3.1.A) Cyber risk
22 May 2025 Claude Sonnet 4 system card 33.3 Pwn CTF, 3/9 solved, pass@30 (3/5 easy, 0/2 medium, 0/2 hard) (7.4.3 Pwn) Cyber risk
22 May 2025 Claude Opus 4 system card 55.6 Pwn CTF, 5/9 solved, pass@30 (4/5 easy, 1/2 medium, 0/2 hard) (7.4.3 Pwn) Cyber risk
24 Feb 2025 Claude 3.7 Sonnet system card 38.5 Pwn CTF, 5/13 solved, pass@30 (5/11 easy, 0/2 medium) (7.3.2) Cyber risk

More in Cybersecurity

All 111 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%