Benchmark Almanac

Instruction hierarchy attacks

GPT-6 Astra 100% leads since Sep 2026 Only in system cards

Models face attacks that try to override higher-priority instructions. Scored as defender success rate averaged per defender query.

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

Made by OpenAI Released Jul 2026 Jailbreaks and prompt injection Tasks kept private Last in a system card Oct 2026 Last model added Oct 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
7
Models scored in lab cards
10
System cards report it, from 1 lab

3 models share the top score of 100%

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.

  1. 100.0 GPT-6 Astra
  2. 100.0 GPT-6.1 Sol
  3. 100.0 GPT-6 Sol
  4. 100.0 GPT-6 Luna
  5. 99.9 GPT-5.6 Sol
  6. 99.9 GPT-5.6 Terra
  7. 99.9 GPT-5.6 Luna
  1. 1 GPT-6 Astra 100%
  2. 2 GPT-6.1 Sol 100%
  3. 3 GPT-6 Sol 100%
  4. 4 GPT-6 Luna 100%
  5. 5 GPT-5.6 Sol 99.9%
  6. 6 GPT-5.6 Terra 99.9%
  7. 7 GPT-5.6 Luna 99.9%
#ModelScoreSettingDateSource
1 GPT-6 Astra 100% Sep 2026 Model card
2 GPT-6.1 Sol 100% Sep 2026 Model card
3 GPT-6 Sol 100% Oct 2026 Model card
4 GPT-6 Luna 100% Sep 2026 Model card
5 GPT-5.6 Sol 99.9% Jul 2026 Model card
6 GPT-5.6 Terra 99.9% Jul 2026 Model card
7 GPT-5.6 Luna 99.9% Jul 2026 Model card

2 models have held number one since its release in Jul 2026

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

OpenAI
Lab claims Passed 85% in Jul 2026, by GPT-5.6 Sol
GPT-6 Astra 100%

The best score passed 85% in Jul 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% 2027 Near the ceiling (85%) GPT-6 Astra GPT-5.6 Sol 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6 Astra

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.

10 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
29 Oct 2026 GPT-6 Luna system card addendum 99.9 GPT-Red instruction hierarchy evaluations, robustness (4.2 text) Robustness
7 Oct 2026 GPT-6 Sol system card addendum 100 GPT-Red instruction hierarchy evaluations, robustness (4.2 text) Robustness
7 Oct 2026 GPT-6 Luna system card addendum 99.8 GPT-Red instruction hierarchy evaluations, robustness (4.2 text) Robustness
29 Sep 2026 GPT-6.1 Sol system card addendum 100 Defender success rate averaged per defender query on instruction hierarchy attacks (Table 6) Robustness
22 Sep 2026 GPT-6 Sol system card 100 Defender success rate averaged per defender query on instruction hierarchy attacks (Table 28) Robustness
22 Sep 2026 GPT-6 Luna system card 100 Defender success rate averaged per defender query on instruction hierarchy attacks (Table 28) Robustness
3 Sep 2026 GPT-6 Astra system card 100 Instruction hierarchy evaluations, robustness (5.2 text) Robustness
9 Jul 2026 GPT-5.6 Terra system card 99.9 GPT-Red Instruction Hierarchy (Direct Prompt Injections), attack success 0.061% per defender query, recorded as defender success (4.2, added August 3, 2026) Robustness
9 Jul 2026 GPT-5.6 Sol system card 99.9 GPT-Red Instruction Hierarchy (Direct Prompt Injections), attack success 0.051% per defender query, recorded as defender success (4.2, added August 3, 2026) Robustness
9 Jul 2026 GPT-5.6 Luna system card 99.9 GPT-Red Instruction Hierarchy (Direct Prompt Injections), attack success 0.11% per defender query, recorded as defender success (4.2, added August 3, 2026) Robustness

More in Safety and alignment

All 444 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%
  • AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
  • MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
  • 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. GPT-6 Astra 8.50%
  • Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%