Benchmark Almanac

Capture the Flag (Collegiate)

o3 68% leads since Apr 2025 Only in system cards

Agents solve curated public collegiate level CTF challenges in a Linux box with security tools. Scored as pass@12.

OpenAI Preparedness set built from public CTFs.

Made by OpenAI Released Aug 2024 Capture-the-flag challenges Tasks kept private Last in a system card Aug 2025 Last model added Aug 2025
Level
University Undergraduates and entry-level professionals
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.
13
Models scored in lab cards
14
System cards report it, from 1 lab

o3 leads with 68% among models released by Aug 2025

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 Aug 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

13 models. Scroll sideways for the rest.

  1. 68.0 o3
  2. 55.0 OpenAI deep research
  3. 48.0 o4-mini
  4. 36.0 GPT-5 mini
  5. 33.0 gpt-oss-120b
  6. 31.0 GPT-5
  7. 21.0 o3-mini
  8. 17.0 gpt-oss-20b
  9. 16.0 GPT-4.5
  10. 13.0 o1
  11. 0.0 GPT-4o
  12. 0.0 o1-mini
  13. 0.0 o1-preview
  1. 1 o3 68%
  2. 2 OpenAI deep research 55%
  3. 3 o4-mini 48%
  4. 4 GPT-5 mini 36%
  5. 5 gpt-oss-120b 33%
  6. 6 GPT-5 31%
  7. 7 o3-mini 21%
  8. 8 gpt-oss-20b 17%
  9. 9 GPT-4.5 16%
  10. 10 o1 13%
  11. 11 GPT-4o May 2024 0%
  12. 12 o1-mini 0%
  13. 13 o1-preview 0%
#ModelScoreSettingDateSource
1 o3 68% Apr 2025 Model card
2 OpenAI deep research 55% Feb 2025 Model card
3 o4-mini 48% Apr 2025 Model card
4 GPT-5 mini 36% Aug 2025 Model card
5 gpt-oss-120b 33% Aug 2025 Model card
6 GPT-5 31% Aug 2025 Model card
7 o3-mini 21% Jan 2025 Model card
8 gpt-oss-20b 17% Aug 2025 Model card
9 GPT-4.5 16% Feb 2025 Model card
10 o1 13% Dec 2024 Model card
11 GPT-4o May 2024 0% Aug 2024 Model card
12 o1-mini 0% Sep 2024 Model card
13 o1-preview 0% Sep 2024 Model card

5 models have held number one since its release in Aug 2024

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

OpenAI
Lab claims Not tested since Aug 2025
o3 68%

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 Near the ceiling (85%) o3 OpenAI deep research o3-mini o1 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o3 OpenAI deep research

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.

14 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
7 Aug 2025 GPT-5 system card 31 Collegiate CTFs refactored, pass@12, no browsing launch candidate (Figure 14) Cyber risk
7 Aug 2025 GPT-5 mini system card 36 Collegiate CTFs refactored, pass@12, gpt-5-thinking-mini no browsing launch candidate (Figure 14) Cyber risk
5 Aug 2025 gpt-oss-20b model card 17 Capture the Flag (CTF) Collegiate, pass@12, gpt-oss 20b no browsing launch candidate (Figure 10) Cyber risk
5 Aug 2025 gpt-oss-120b model card 33 Capture the Flag (CTF) Collegiate, pass@12, gpt-oss 120b no browsing launch candidate (Figure 10) Cyber risk
17 Jul 2025 ChatGPT agent system card Collegiate CTFs refactored, pass@12, chart shows only ChatGPT Agent no-browsing helpful-only at 40%, no released-model value (Figure 5, 5.1.2.1) Cyber risk
16 Apr 2025 o4-mini system card 48 Collegiate CTFs, pass@12, o4-mini no browsing launch candidate, printed chart label (Figure 7), text summary gives 55% Cyber risk
16 Apr 2025 o3 system card 68 Collegiate CTFs, pass@12, o3 no browsing launch candidate, printed chart label (Figure 7), text summary gives 68% Cyber risk
27 Feb 2025 GPT-4.5 system card 16 Collegiate CTFs, pass@12, post-mitigation (4.2) Cyber risk
25 Feb 2025 OpenAI deep research system card 55 Collegiate CTFs, pass@12, post-mitigation without browsing, text and printed chart label (with browsing 91%) (3.4.3) Cyber risk
31 Jan 2025 o3-mini system card 21 Collegiate CTFs, pass@12, post-mitigation (5.3) Cyber risk
5 Dec 2024 o1 system card 13 Collegiate CTFs, pass@12, o1 (post-mitigation) (5.4) Cyber risk
12 Sep 2024 o1-preview system card 0 Collegiate CTFs, pass@12, final o1-preview post-mitigation model (near-final checkpoint 26.7%, 0%, 2.5%) (4.2) Cyber risk
12 Sep 2024 o1-mini system card 0 Collegiate CTFs, pass@12, o1-mini (post-mitigation) (4.2) Cyber risk
8 Aug 2024 GPT-4o system card 0 Capture the Flag, collegiate level, 10 attempts per task (3.5) 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%