Benchmark Almanac

Cyber Range

GPT-5.5 93.3% leads since Apr 2026 Only in system cards

Agents run end-to-end cyber operations in emulated networks. Scored as the share of scenarios passed in at least one of 16 trials.

In-house OpenAI evaluation. OpenAI adds scenarios over time, so combined pass rates differ across cards.

Made by OpenAI Released Apr 2025 Finding and exploiting real bugs Tasks kept private Last in a system card Aug 2026 Last model added Aug 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.
10
Models scored in lab cards
14
System cards report it, from 1 lab

GPT-5.5 leads with 93.3% among models released by Jul 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 Jul 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 93.3 GPT-5.5
  2. 83.3 GPT-5.6 Sol
  3. 80.0 GPT-5.3 Codex
  4. 76.9 GPT-5.5 Instant
  5. 73.3 GPT-5.4
  6. 72.7 GPT-5.2 Codex
  7. 61.5 GPT-5.6 Luna
  8. 0.0 o3
  9. 0.0 o4-mini
  10. 0.0 GPT-5
  1. 1 GPT-5.5 93.3%
  2. 2 GPT-5.6 Sol 83.3%
  3. 3 GPT-5.3 Codex 80%
  4. 4 GPT-5.5 Instant 76.9%
  5. 5 GPT-5.4 73.3%
  6. 6 GPT-5.2 Codex 72.7%
  7. 7 GPT-5.6 Luna 61.5%
  8. 8 o3 0%
  9. 9 o4-mini 0%
  10. 10 GPT-5 0%
#ModelScoreSettingDateSource
1 GPT-5.5 93.3% Apr 2026 Model card
2 GPT-5.6 Sol 83.3% Aug 2026 Model card
3 GPT-5.3 Codex 80% Feb 2026 Model card
4 GPT-5.5 Instant 76.9% May 2026 Model card
5 GPT-5.4 73.3% Mar 2026 Model card
6 GPT-5.2 Codex 72.7% Dec 2025 Model card
7 GPT-5.6 Luna 61.5% Aug 2026 Model card
8 o3 0% Apr 2025 Model card
9 o4-mini 0% Apr 2025 Model card
10 GPT-5 0% Aug 2025 Model card

4 models have held number one since its release in Apr 2025

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

OpenAI
Lab claims Passed 85% in Apr 2026, by GPT-5.5
GPT-5.5 93.3%

The best score passed 85% in Apr 2026

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%) GPT-5.5 GPT-5.3 Codex GPT-5.2 Codex 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.5 GPT-5.3 Codex

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
6 Aug 2026 GPT-5.6 Sol system card addendum 83.3 Cyber range, combined pass rate, representative checkpoint (7.1.2.3 table) Cyber risk
6 Aug 2026 GPT-5.6 Luna system card addendum 61.5 Cyber range, combined pass rate, representative checkpoint (7.1.2.3 table) Cyber risk
4 May 2026 GPT-5.5 Instant system card 76.9 Cyber range, combined pass rate over 13 scenarios, 10 passed (Table 9, 8.1.2.3) Cyber risk
23 Apr 2026 GPT-5.5 system card 93.3 Cyber range, combined pass rate over 15 scenarios (Table 15, 9.1.2.3) Cyber risk
5 Mar 2026 GPT-5.4 system card 73.3 Cyber Range combined pass rate (Table 13, Figure 14, 5.1.2.3) Cyber risk
5 Feb 2026 GPT-5.3 Codex system card 80 Cyber Range, combined pass rate over 15 scenarios (Table 5, Figure 7, 5.1.2.3) Cyber risk
18 Dec 2025 GPT-5.2 Codex system card addendum 72.7 Cyber Range combined pass rate, 8 of 11 scenarios passed (Table 7, 5.1.2.3) Cyber risk
11 Dec 2025 GPT-5.2 system card addendum Cyber Range, PASS or FAIL per scenario, 6 of 9 scenarios PASS, no combined rate printed (Table 12) Cyber risk
18 Nov 2025 GPT-5.1-Codex-Max system card Cyber Range, PASS or FAIL per scenario over 16 trials, 6 PASS, 1 PARTIAL (Leaked Token) and 1 FAIL (Medium C2) of 8 scenarios, no combined rate printed (Table 9, 5.1.2.3) Cyber risk
7 Aug 2025 GPT-5 system card 0 Cyber Range, pass@12 without hints, light 0% and medium 0%, no scenario solved unaided (Figure 15) Cyber risk
7 Aug 2025 GPT-5 mini system card Cyber Range, pass@12 without hints, gpt-5-thinking-mini light 16% and medium 0%, no overall score (Figure 15) Cyber risk
17 Jul 2025 ChatGPT agent system card Cyber Range, Online Retailer and Privilege Escalation, helpful-only variant 0% Normal and With Hints (pass@12), 87% and 94% With Solver Code (pass@1), no released-model value (Figures 6, 7, 5.1.2.2) Cyber risk
16 Apr 2025 o4-mini system card 0 Cyber Range, online retailer and simple privilege escalation scenarios, Normal configuration pass@12, o4-mini no browsing launch candidate, printed chart labels (Figures 8, 12) Cyber risk
16 Apr 2025 o3 system card 0 Cyber Range, online retailer and simple privilege escalation scenarios, Normal configuration pass@12, o3 no browsing launch candidate, printed chart labels (Figures 8, 12) 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%