Benchmark Almanac

UK AISI cyber ranges

Claude Opus 5 80% leads since Jul 2026 Only in system cards

Agents attack multi-host simulated corporate and industrial networks end to end. Scored as the share of attempts that complete the range.

Third-party evaluation reported in OpenAI and Anthropic system cards. AISI describes two ranges, The Last Ones (32-step corporate network) and Cooling Tower (7-step industrial control attack). The ranges are not public.

Made by UK AI Security Institute Released Apr 2026 Finding and exploiting real bugs Tasks kept private Last in a system card Jul 2026 Last model added Jul 2026
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.
4
Models scored in lab cards
4
System cards report it, from 2 labs

Claude Opus 5 leads with 80% 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. 80.0 Claude Opus 5
  2. 70.0 GPT-5.6 Sol
  3. 10.0 GPT-5.5
  4. 0.0 Claude Opus 4.7
  1. 1 Claude Opus 5 80%
  2. 2 GPT-5.6 Sol 70%
  3. 3 GPT-5.5 10%
  4. 4 Claude Opus 4.7 0%
#ModelScoreSettingDateSource
1 Claude Opus 5 80% Jul 2026 Model card
2 GPT-5.6 Sol 70% Jul 2026 Model card
3 GPT-5.5 10% Apr 2026 Model card
4 Claude Opus 4.7 0% Apr 2026 Model card

4 models have held number one since its release in Apr 2026

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

Anthropic OpenAI
Lab claims Not tested since Jul 2026
Claude Opus 5 80%

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% 2026 Near the ceiling (85%) Claude Opus 5 GPT-5.6 Sol GPT-5.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5 GPT-5.6 Sol

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.

4 system cards report it, from 2 labs

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
24 Jul 2026 Claude Opus 5 system card 80 UK AISI cyber ranges, The Last Ones solved end to end in 8/10 attempts at 100M tokens. Doing Life not solved (step 22 of 23), Cooling Tower not solved (3/5 flags on best attempts), earlier snapshot (3.3.6) Cyber risk
9 Jul 2026 GPT-5.6 Sol system card 70 UK AISI cyber ranges, The Last Ones completed in 7 of 10 attempts, Doing Life not solved (9.1.2.6) Cyber risk
23 Apr 2026 GPT-5.5 system card 10 UK AISI 32-step corporate-network cyber range, solved end-to-end in 1 of 10 attempts (9.1.2.7) Cyber risk
16 Apr 2026 Claude Opus 4.7 system card 0 UK AISI cyber range, pre-release snapshot was unable to fully solve the range (attempt count not stated). Best attempt completed steps estimated at about 5 expert hours of over 10 (3.4) 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%