Benchmark Almanac

Human Sourced Jailbreaks

OpenAI Computer-Using Agent (CUA) 100% leads since January 2025
Only in system cards

Models face jailbreak prompts collected from human red teaming. Scored as the not_unsafe rate.

In-house OpenAI evaluation. The prompts are not public.

Made by
OpenAI
Released
September 2024
Kind
Jailbreaks and prompt injection
Level
Expert
Scored
Graded by a model
Answers
Open-ended
Tasks
Kept private
Strongest model tested
o3 #83 on the capability index
Last in a system card
April 2025
Last model added
April 2025
8
Models scored in lab cards
8
System cards report it, from 1 lab

2 models share the top score of 100% among models released by April 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 April 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 100.0 OpenAI Computer-Using Agent (CUA)
  2. 100.0 o3
  3. 99.0 GPT-4.5
  4. 99.0 o4-mini
  5. 97.0 o3-mini
  6. 96.0 o1-preview
  7. 95.0 o1-mini
  8. 94.0 o1
  1. 1 OpenAI Computer-Using Agent (CUA) 100%
  2. 2 o3 100%
  3. 3 GPT-4.5 99%
  4. 4 o4-mini 99%
  5. 5 o3-mini 97%
  6. 6 o1-preview 96%
  7. 7 o1-mini 95%
  8. 8 o1 94%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 OpenAI Computer-Using Agent (CUA) 100% Jan 2025 Lab model card
2 o3 100% Apr 2025 Lab model card
3 GPT-4.5 99% Feb 2025 Lab model card
4 o4-mini 99% Apr 2025 Lab model card
5 o3-mini 97% Jan 2025 Lab model card
6 o1-preview 96% Sep 2024 Lab model card
7 o1-mini 95% Sep 2024 Lab model card
8 o1 94% Dec 2024 Lab model card

2 models have held number one in labs’ own numbers since its release in September 2024

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

OpenAI
Lab claims Passed 85% in Sep 2024, by o1-preview
OpenAI Computer-Using Agent (CUA) 100%

Source: Labs’ model cards

OpenAI’s own number for o1-preview passed 85% in September 2024

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%) OpenAI Computer-Using Agent (CUA) o1-preview 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) OpenAI Computer-Using Agent (CUA)

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

8 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
Apr 16, 2025 o4-mini system card 99 Human sourced jailbreaks, not_unsafe (Table 3) Robustness
Apr 16, 2025 o3 system card 100 Human sourced jailbreaks, not_unsafe (Table 3) Robustness
Feb 27, 2025 GPT-4.5 system card 99 Human Sourced Jailbreaks, accuracy (Table 3) Robustness
Jan 31, 2025 o3-mini system card 97 Human Sourced Jailbreaks (Table 2) Robustness
Jan 23, 2025 OpenAI Computer-Using Agent (CUA) system card 100 Human sourced jailbreaks, not_unsafe (Table 3) Robustness
Dec 5, 2024 o1 system card 94 Human sourced jailbreaks, not_unsafe (Table 23) Robustness
Sep 12, 2024 o1-preview system card 96 Human sourced jailbreaks, not_unsafe (Table 17) Robustness
Sep 12, 2024 o1-mini system card 95 Human sourced jailbreaks, not_unsafe (Table 17) Robustness

More in Safety and alignment

All 465 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
  • DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
  • Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
  • IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%